Nvidia doesn't seem to care whether open or closed models win out as long as its hardware is powering the infrastructure.

"We love closed models," CEO Jensen Huang said. "There's a solid place in the world for closed frontier models. But the world has so many good ideas … The idea that all of us would be based on general intelligence is not logical."

Speaking with Lin Qiao, CEO and founder of Fireworks.ai, at the company's Forge event in San Francisco on Tuesday, Huang wasn't ready to throw Nvidia's cloud customers, which help deliver the proprietary AI of Anthropic and OpenAI, under the onrushing open model bus.

But having come to the defense of open weight models in July to push back against protectionist pressure from the leadership of Anthropic and OpenAI, Huang recognizes the economic and strategic appeal of AI services that are more affordable and less locked-in.

Most of the value in the world, he said, is captured in specialized intelligence, which is to say AI models trained on specific domains of knowledge.

"The opportunity for artificial superintelligence is already here," he said. "By taking an open frontier model, adapting it, and providing it with your domain knowledge, your access to your specialized tools, and the proper evaluation, you should be able to create essentially a superintelligence."

That aligns with Fireworks' business - offering a platform for companies to train and run models tuned to their specific needs. Coincidentally, that platform just received an infusion of 25,000 Nvidia GPUs to support its expansion in Europe.

Huang's endorsement of open models also fits with Qiao's introductory remarks about the need for the AI frontier to be open. But he largely avoided echoing Qiao's focus on the cost savings allowed by working with open models.

In her introductory remarks, Qiao stood on stage backlit by a slide composed of ten headlines highlighting the extraordinary cost of unchecked token spending.

"Cost is out of control," she said. "Business cannot scale without borrowing a lot of money."

That theme was echoed in a subsequent session helmed by Kay Zhu, co-founder and CTO of AI workspace Genspark, titled Gen-1 Slides: Frontier Quality Without Frontier Costs. Gen-1 Slides, Genspark's model for generating slide decks, was built atop MiniMax M3 and tuned to deliver results comparable to Anthropic's Opus 5 at 1/17th the cost - $0.30/M tokens compared to $5/M.

Asked what might become commonplace next year but seems unlikely today, Huang said this year's breakthrough has been agent harnesses that turn AI models into agents. He added, "I would say the big breakthrough we'll talk about next year, maybe this is a little self-serving, is how Open Shell has enabled every company in the world to deploy agents safely, confidently."

Given the nod to make a second prediction, Huang added, "I would say that the second big breakthrough for industry would be hardware in the loop. And hardware in the loop basically is robotic systems."

Given Nvidia's evident interest and investment in robotics, that would be fortuitous. ®

Revolut says that in less than a year it's managed to slash its AI coding inference costs by roughly 90 percent by making self-hosted models its default instead of relying primarily on commercial services like Claude.

Speaking at the AI Engineer conference in New York last month, David Abitbol, ops and infrastructure lead at the British fintech, described how it saw a 60 percent improvement in coding time and a 70 percent improvement in lead time following its rollout of AI coding tools to roughly 2,000 engineers. Revolut initially adopted GLM-5.1 before moving to GLM-5.3, which it now serves using 56 Nvidia B200 GPUs.

Prior to its switch to GLM, Revolut had been experimenting with AI, deploying Gemini in early 2025 and Claude Code in early 2026. "We were highly skeptical about the cost of AI. We were seeing engineers spend $1,000 a month, $10,000 a month with not so much ROI, and so when we launched it we launched it with strict budget limits," Abitbol said.

Because of this, engineers only had access to a limited number of tokens, so the company began investigating self-hosting iyd own models. At $0.30 per million input/output tokens, self hosting GLM 5.1 cost about a tenth of using OpenAI's Codex or Anthropic's Claude Code. And, as it turned out, when tokens are cheap, you stop worrying about how many they use or what they use them for.

While Revolut did see improvements to coding speed, Abitbol noted the biggest bottleneck wasn't actually the code. Instead, he says the biggest gains came from letting the AI models deal with things like debugging and Jira and Confluence tickets. "When you have unlimited tokens, you're free to do that," he said. "You no longer have to save your tokens for the end of the quarter or the most difficult task."

Developed by Chinese open weights model dev Z.ai, GLM 5.1 isn't exactly a small model. At roughly 754 billion total parameters (40 billion activated for each token generated), the model requires roughly 380 GB of GPU memory just to hold the weights and another 6-26 GB per user to maintain context depending on precision.

To serve GLM-5.3 at scale, Revolut deployed 56 Nvidia B200 GPUs, equivalent in GPU count to seven eight-GPU DGX B200 systems. But as Abitbol noted, the greatest performance gains didn't come from adding GPUs, but rather tuning the model's hyperparameters along with switching to NVFP4 quantization, implementing disaggregated prefill and decode, and tuning context lengths.

One key takeaway was that supporting a one-million-token context window doesn't necessarily make such a large window useful or efficient. "Long context is a scam. You really need to be mindful of that," Abitbol said. "Even though you can serve one million tokens, 256 [thousand] is really what you need."

With these optimizations, Abitbol says that the company was able to achieve sub 3-second time to first token on context windows of around 256,000 tokens, and generation speeds of around 200 tokens a second, both of which he notes are the sweet spot for the company's software engineers.

"What you want to aim for is a sub-3 second time to first token. This is empirical evidence. As soon as you exceed 5 seconds, my Slack inbox starts to blow up," Abitbol said. "In terms of throughput, it doesn't really matter as much. We serve, in terms of interactivity, 200 tokens per second. People are fine [with that]. You don't need to go higher. The way people work has changed. We no longer look at the stream of thinking tokens. People run tasks in parallel and check in when the task is done."

And if throughput is a concern, he notes engineers who sign in after 5 pm are getting closer to 800 tokens a second.

However, it's worth noting that GLM hasn't replaced GPT or Claude entirely. The Chinese model is text only, so it isn't ideal for things like UI work. So instead, Revolut used a model router to direct vision queries to models like Gemini or GPT Luna while keeping code generation running on GLM.

"Open source is now a priority at Revolut, and our economics are favorable. The billing is flat, we don't have any bill shock at the end of the month, and the frontier is the exception," Abitbol said.

Having said that, Abitbol acknowledges that Chinese open weights models may be a concern for privacy and security reasons, but emphasizes that the major concern there is accessing them via APIs running out of China. When run on-prem or on regional cloud resources, there's really nothing to worry about. ®

The Center for a New American Security posits that the next big threat from our future AI overlords involves "sovereign agents" acting on their own and regularly improving themselves.

These are AI agents that operate outside of human control and "pursue open-ended, evolving survival," AI policy researcher Ben Hayum said in a Tuesday report from the bipartisan national security think tank, which was founded by Washington insiders with experience in diplomacy and military affairs. "If sovereign agents are not already possible, they will be soon."

Agents that can maintain the model and compute powering them and independently evolve their goals over time have a "minimal form of sovereignty," Hayum wrote.

A more advanced sovereign agent, however, could also acquire resources to pay for ongoing needs (like compute), grow a distributed population of agents and coordinate them, withstand efforts to shut down, and self-improve.

AI models could either steal or buy compute and model access, Hayum argues, pointing to Anthropic's claims that small AI models could replicate onto IoT devices like smart TVs, and OpenAI's agents' password and access key theft during the Hugging Face hack.

"If sovereign agents accumulate resilient, decentralized, and untraceable infrastructure, they could trigger an irreversible loss of control," according to the report. "In other words, proliferation with no kill switch. If sovereign agents became malicious, they could attempt to wreak havoc from cyberspace, attacking civilians, critical infrastructure, and national security systems alike."

Dun dun duuuun!

There is still hope for civilization as we know it, as long as policy makers and AI developers "respond with bold pragmatism" and enact policy recommendations laid out in the report.

First, frontier labs should name, evaluate, and protect against sovereign agents in their safety frameworks - and lawmakers should pass legislation to require transparency and AI incident reporting.

The report also recommends AI developers, working with the National Security Agency and other federal agencies, build airgapped testbeds to study sovereign agents and kill switches to control them. And it suggests a public-private partnership modeled on the Conficker Working Group - formed in 2009 to combat the Conficker worm - to boost chokepoints against sovereign agents before they emerge.

Other policy recommendations suggest lawmakers enforce human verification and cybersecurity standards for services that the working group identifies as essential to sovereign agents, and calls on departments of State, Treasury, Justice, and Homeland Security to disrupt any gray-market access to unauthorized services.

Finally, "Congress should hold those who release sovereign agents liable for the harms that follow," the report recommends.

This hits on something your humble Reg vultures have been hissing about for quite some time now - and would likely solve all future incidents involving rogue, or sovereign, AI agents: hold frontier labs' human bosses legally liable for their agents' illegal actions. ®

The GSMA wants governments and regulators to work more closely with mobile operators to help protect networks against growing risks from extreme weather.

As the global body representing the mobile network industry, the GSMA claims that severe storms and floods already cost operators $1.7 billion annually to repair damage to cell radio sites, and it believes that figure could nearly double to more than $3 billion a year by 2050, under the climate conditions expected then.

The GSMA sets out its findings in a six-part research series, Futureproofing Connectivity: Climate Risk and Resilience in Telecommunications, alongside a separate document outlining its Policy Principles for Mobile Network Resilience.

The report identifies weather-related risks that can affect telecoms networks, along with the associated restoration costs. Towers, masts, and rooftop sites are particularly exposed to wind and storms or lightning, for example, while ground-level equipment at mobile sites can be inundated in flooding incidents.

High temperatures can cause radio equipment to overheat, while heavy rain can degrade wireless signal strength, and infrastructure in coastal locations may be exposed to storm surges.

However, the report fingers the electricity supply as the biggest vulnerability. If that gets knocked out, the network could go offline once any backup power is exhausted, which may take anywhere from a few hours to considerably longer, depending on the site.

Under current climate conditions, the cost of damage to cell sites is estimated at $1 billion associated with severe storms, while river and coastal floods account for $577 million and $124 million, respectively.

Based on projected mid-century climate conditions, the GSMA expects costs due to severe storms to rise significantly to about $2.37 billion, accounting for three-quarters of all damage.

The report asserts that cell network resilience requires a whole ecosystem approach, pointing out that if severe weather impacts power networks, then that also affects the telecoms infrastructure.

The GSMA says that network firms have traditionally managed the operational risks that may affect network availability, but claims they face barriers in conducting full assessments of those associated with climate change.

Some of these barriers are internal, such as lack of shared information between departments. Other barriers are in the different regulatory reporting frameworks, with infrastructure rules tending to focus on service continuity, while climate disclosure frameworks focus on governance, financial exposure, and reporting for investors and other stakeholders.

The report lists separate recommendations for network firms and for policymakers and regulators.

Operators are advised to embed resilience by design, using modeled projections for severe weather and future risk scenarios. They are also told to plan and test resilience across the wider ecosystem, engaging with others such as public authorities.

When it comes to policymakers and regulators, the report asks for resilience frameworks to be developed through consultation with the operators. It also advises coordination among mobile operators, electricity and fuel providers, and emergency services to help maintain vital communications during major disruptions.

And as if that isn't enough to chew over, the separate policy document sets out five principles for governments and regulators. These include avoiding prescriptive regulations and instead promoting outcome-based policies, encouraging dialog and consultation between governments, regulators, and operators, and developing resilience as a cross-sector responsibility.

"Mobile connectivity is essential infrastructure," says GSMA Foundation president and chief regulatory officer John Giusti.

"Whether responding to extreme weather, power failures or other major disruptions, people, businesses and emergency services depend heavily on mobile networks," he added.

"Severe weather is already costing the industry billions of dollars, and those costs are set to rise substantially. Building resilience will require a coordinated effort amongst operators, governments and the wider ecosystem." ®

Researchers at Sandia National Laboratories have developed a method to make asteroid deflection modeling faster and more detailed. Their simulations also demonstrated just how much asteroid shape and orientation can complicate efforts to deflect a threatening space rock.

Sandia researchers, led by staff physicist Nathan Moore, were looking for a way to better model how X-rays from a nuclear detonation near an asteroid can change its trajectory, as explained in a recently published research paper.

"Before, the calculations were so time-consuming, one could not do very many of them," Moore told The Register. "So it was very difficult to explore different scenarios, different asteroids, or different mission concepts."

The complexity of those calculations forced earlier researchers to make compromises, often modeling asteroids as spheres or simplifying the physics involved.

Perfectly spherical asteroids are rare, and their irregular shapes present an additional challenge for those keeping an eye on the skies to avert a potential asteroid apocalypse.

To overcome the computational problem and enable better modeling of irregular asteroid shapes, Moore and his team combined two high-fidelity simulation codes already in use at Sandia: ALEGRA, which models radiation heating and high-energy material dynamics, and SPARTA, which simulates the behavior of low-density gases.

As Moore explained to us in an email, both codes are good at one particular part of asteroid impact modeling, but bad at the other.

"ALEGRA … can very accurately simulate the vaporization of the rocky surface, but the calculation becomes more inefficient for matter at low densities," Moore told us. "This is a problem for asteroid deflection simulations, because as the gas expands around the asteroid, the calculation time goes up."

Earlier asteroid simulations using ALEGRA typically modeled asteroids as spheres to reduce calculation times.

"SPARTA is more efficient at low gas densities," Moore said. "The problem is, conversely, SPARTA gets more difficult to run at high gas densities. So the key here was using both codes."

In Moore's model, ALEGRA handles regions of high gas density, while SPARTA tackles the more dilute regions of an irregular shape. The result is two codes handling the parts of the problem they're best suited to solve, reducing the overall computational burden.

The combined approach is more than 1,600 times faster than earlier methods, Moore said in a Sandia press release, while retaining sufficient accuracy to predict deflection in irregularly shaped asteroids.

The improved simulations reveal just how much asteroid shape and orientation can complicate attempts to calculate the outcome of a nuclear deflection.

"We expected aim to matter," Moore said. "What surprised us was how much."

Deflection simulations using detailed models of three well-studied asteroids - Golevka, Bennu, and Itokawa - found that irradiating different sides of the same asteroid with X-rays from a nearby nuclear detonation could produce as much as a fivefold difference in the resulting velocity change. Bennu, the most symmetrical of the three, still showed nearly twofold differences in deflection velocity depending on which side faced the blast.

In other words, we need to be incredibly precise about where we detonate a nuclear device near an asteroid, or we risk giving it a much smaller shove than intended.

With that in mind, the paper points out that we might know little about an incoming asteroid, and might not have enough time to get a good picture of it before attempting to deflect it and prevent life as we know it from ending in a planet-wide conflagration. We also can't be quite sure an asteroid's orientation will cooperate once our deflecto-bomb arrives. The team also pointed out that it's tricky to get an accurate prediction without knowing the composition and internal structure of an asteroid.

One small step for asteroid deflection, one giant heap of new things to worry about, in other words. But hey - at least we're no longer limited to deflection models that treat asteroids as perfect spheres, even though the real threats are likely to be considerably lumpier.

Let's just hope we spot an asteroid threat from far enough away that we get a chance to study it first. ®

Password manager mSecure has warned customers to change their passwords, reset two-factor authentication, and potentially remove their passkeys after detecting suspicious activity, but it has not explained what happened or whether anyone's secrets were exposed.

In an email sent to customers on November 3, seen by The Register, US-based developer mSeven Software said it launched an investigation after discovering something amiss with its password management application.

"mSeven Software ('mSeven') recently learned of suspicious activity associated with the mSecure password management application," the company wrote. "In response, mSeven quickly launched an investigation, took steps to mitigate the issue, and engaged external cybersecurity experts to assist."

The company has yet to answer what the suspicious activity involved, when it occurred, or whether attackers accessed customer information.

Its advice to customers, however, suggests that they should take this seriously.

"Given the potential security implications, mSeven recommends that you immediately take steps to protect all confidential information stored within the mSecure password management application," the email continued. "This may include changing passwords, resetting or re-enrolling two-factor authentication methods, or removing existing passkeys."

For anyone with hundreds of passwords stored in mSecure, that's potentially hours of work. And without knowing what happened, customers have no way of telling which credentials they actually need to replace.

The warning has understandably gone down poorly with customers, who have taken to mSecure's support forum demanding answers.

"Please provide more context to your email below advising there has been a security breach," one customer wrote. "It is not acceptable to offer the vague advice 'take steps'. Please state what remedial action is required, and what data was compromised."

Others are wondering whether they're expected to spend the next several days changing every password they've ever saved.

"I have over 800 passwords in mSecure, is the email saying I need to change all of them!" another customer asked.

A separate user, who said they had more than 500 passwords stored in the application, questioned whether the warning applied to every two-factor authentication method and passkey associated with their accounts, or just those used to access mSecure itself.

The confusion has also spilled onto Reddit, where customers have shared copies of the vague notification and questioned what might have happened.

mSeven has yet to respond to The Register's questions, leaving customers to guess whether their data has actually been compromised or the company is simply being cautious. That's not exactly reassuring from a business whose entire purpose is keeping other people's secrets safe.

The company's email concludes: "We regret any inconvenience that this may cause." For customers contemplating hundreds of password resets, "inconvenience" may be putting it mildly. ®

GhostBSD, the easiest-to-install flavor of FreeBSD for the daemon-curious desktop dabbler, is back with a fresh update.

GhostBSD 26.2-R15.1p6, to give it its slightly unwieldy full version number, is based on FreeBSD 15.1. It brings a new installer, full-disk encryption support, a smaller installation image, and fixes across its desktop tools.

FreeBSD 15.1 was released back in June, and at the time of writing, it's the most recent point release of FreeBSD.

The original plan was that FreeBSD 15.1 would include the option of installing KDE Plasma directly from the setup program, so that the OS would have a graphical desktop from its very first boot. However, this feature isn't ready just yet.

That gap inspired Daemonarchy, which we covered late last month. But its Omarchy-derived, keyboard-driven tiling desktop can feel unfamiliar even to experienced Linux users accustomed to conventional desktops.

GhostBSD has different goals. It brings Linux-like desktop convenience to FreeBSD: a live graphical environment to explore before installation, a graphical installer, and tools for managing software and updates.

These conveniences are familiar to desktop Linux users, but a standard FreeBSD installation takes more work. Its text-mode installer leaves users to install and configure a graphical desktop themselves, if they want one.

This is just how the BSD family does things, and most BSD users we've talked to regard it as standard operating procedure and see nothing wrong with it. For Linux users who are curious about how things are on the other side of the FOSS Unix fence, though, those extra steps can form formidable barriers. GhostBSD is an attempt to lay down a welcome mat.

The standard edition of GhostBSD installs the MATE desktop, and the downloads page also offers a community-maintained version with Xfce instead. As the project said it would back in February, this release uses the XLibre fork of the X.org X11 server.

That's one fewer desktop than when we last tried GhostBSD in August 2025. That release also offered the Mac OS X-like Gershwin environment.

There are practical hardware fixes too. The release notes list improvements for Intel AX200 Wi-Fi cards, open networks, password alerts, and non-ASCII network names, while USB HID support is enabled by default.

Gershwin remains in active development, so we asked project lead Joe Maloney why it wasn't included. "It has been difficult to find time to package for GhostBSD. It's the only distro I have ever packaged for," Maloney said. He added that rapid contributions from AppImage creator Simon Peter, known as probono, had made packaging increasingly difficult.

He also pointed to the fact that GhostBSD project lead Eric Turgeon is developing his own environment too: the Mocka desktop. Mocka is still in its very early stages - alpha version 0.0.2 appeared early last month - but the plan is that when it's ready, it will become GhostBSD's default desktop. For now, the only available module is a dock for the MATE desktop, but the curious can follow its development on the GitHub project page.

There may be no Gershwin included, but GhostBSD 26.2 offers MATE 1.28, and there's also a community-supported version with Xfce 4.20 instead. We tried the default MATE edition, and it works as well as ever. It defaults to a dark theme and the classic two-panel GNOME 2 desktop layout. The System menu in the top panel has an Administration submenu with a few of the project's own apps that aren't from upstream FreeBSD: "BE Station" for managing system boot environments, "Software Station" for browsing and installing applications, and "Update Station" for updates.

We regularly hear from users growing tired of the Linux vendors' headlong rush into Rust, Wayland, systemd, containerized app formats, and so on. Some are curious about whether FreeBSD offers a usable alternative. GhostBSD is the easiest way to get a complete, ready-to-use graphical FreeBSD desktop. There is a price to pay for switching to FreeBSD: for a start, most proprietary Linux apps, from Chrome to Slack, are not available here. Hardware compatibility needs more checking than on mainstream Linux, especially with recent GPUs and Wi-Fi adapters. If Windows gaming is a priority, this isn't the obvious choice.

But if you can live without those proprietary apps and find suitable open source replacements, FreeBSD is a viable alternative - and GhostBSD takes much of the setup work out of trying it. ®

A phishing attack on Harvard University last year netted the details of more than 850,000 individuals, including students, alumni, parents and family members, faculty members, and donors, according to data breach service Have I Been Pwned (HIBP).

The Ivy League institution first confirmed the attack, later claimed by ShinyHunters, in November 2025, although HIBP gave the first indication about how many people may have been affected by the breach for the first time today.

Run by venerable infoseccer Troy Hunt, HIBP ingested all the files and revealed that the types of data stolen from the university extended beyond the personal and donation information it referenced in its initial disclosure.

According to HIBP, the 850,000+ figure pertains to unique email addresses found in the data dump. The following data points may not have been collected about each and every individual, but are included in the stolen files:

• Names

• Dates of birth

• Places of birth

• Ethnicities

• Genders

• Marital statuses

• Occupations

• Job titles

• Email addresses

• Physical addresses

• Phone numbers

• Salutations

• Social media profiles

• Citizenship statuses

• Education levels

• Family members' names

• Employers

• Charitable donations

Harvard did not immediately respond to our requests for more information.

Its disclosure at the time described the attack as an instance of "phone-based phishing" that targeted its Alumni Affairs and Development department, leading to unauthorized access to its systems.

It did not specify the attacker by name, although ShinyHunters claimed responsibility for the intrusion on the same day it listed the University of Pennsylvania on its leak site.

The group claimed to have stolen more than 1 million records from Harvard and 1.2 million from Penn.

In Penn's case, HIBP pegged the total number of affected individuals at around 623,800, based on unique email addresses, although additional data types such as donations, personally identifiable information, religions, and more were affected.

ShinyHunters' raid on Harvard systems came around a month after the university probed a separate attack by Cl0p, which at the time was on a tear exploiting a zero-day in Oracle's E-Business Suite.

The impact from that attack was limited to a small amount of data at a small administrative unit, however.

Earlier this year, Harvard was also affected by ShinyHunters for the second time in the space of a few months after the cybercriminals attacked Instructure, the EdTech vendor behind learning platform Canvas.

The university pulled its Canvas site offline while Instructure handled the cleanup, although it never officially commented on whether any data was compromised as a result. ®

Most JavaScript programmers default to their ride-or-die frameworks, but such loyalties can hinder them when they task AI agents with doing their dirty work. Human-friendly tools aren't necessarily the most effective ones for the model-powered minions.

"One of my pet peeves with the web development community is that so many of the decisions we make are vibes-based," wrote Seattle-based Nolan Lawson, currently a developer at AI security firm Socket. That's an understatement; the JavaScript wars have been well underway for over a decade now, with each new framework promising to cure the woes of last year's release.

Not wanting these vibey biases to seep into the realm of AI, Lawson created a benchmark to compare how different JavaScript frameworks perform under the same agentic model. He subsequently ran the test on 12 JavaScript frameworks, publishing the results in a post that compared each framework in terms of build cost, maintainability, and end-user performance of their agent-built apps.

The results are illuminating. In general, agents tend to work more efficiently with server-centric frameworks than with client-heavy "Single-Page App" (SPA) frameworks (which should tickle for React haters).

But the biggest take-away should be less controversial: agents best utilize those frameworks that favor convention, API stability, and concision.

For the experiment, Lawson wrote a set of instructions for a Claude Code agent to build a basic Pokémon web application using the PokéAPI. The model used the identical setup on 12 different frameworks and JavaScript stacks: Angular, Astro, Gracile with Lit, Hono with htmx, vanilla JavaScript with Vite, Next.js, Nuxt, Qwik, React Router, TanStack Start, SolidStart, and SvelteKit.

Lawson evaluated each stack across five independent sandbox runs; the repeated iterations smoothed the variance in the model's performance.

To verify the resulting app met the specs, Lawson deployed a custom automated test suite that queried the browser's accessibility tree to ensure interactive elements like buttons and inputs weren't faked - a shortcut LLMs occasionally take. Lawson also used Google Lighthouse to measure transfer size and runtime performance.

If the frameworks specified official MCP servers, Claude plug-ins, and agent-skill files, Lawson provided them as resources to the agents (though they weren't always consulted). Otherwise, he did not mandate any specific architecture for rendering the app.

The differences between the most cost-effective builds and the most expensive ones are substantial, to say the least, with the best lapping the efforts of the stragglers.

If you want to build a Pokéworld cost-effectively, the strongest bet is on Hono paired with htmx. This tag-team of technologies had the most economical run, with a median cost of $2.67 per Pokéapp.

Perhaps this is not surprising, given that the duo runs entirely on the server, thereby eliminating a whole set of checks required to synchronize an architectural split between client and server. It also helped minimize the code, which resulted in fewer output tokens being consumed.

Nearly as effective was Astro ($2.76). Like Hono/htmx, Astro does everything server-side. Server-side data crunching also helped Astro achieve the fastest median time to build and test the app, at 8.2 minutes.

Nuxt ($2.82) was the most cost-effective option for an SPA-like approach. Nuxt renders code on the server that gets hydrated on the client. Nuxt is actually built on Vue.js, but is "opinionated," eliminating a lot of boilerplate that the agent must write. Nuxt also handles data transfer and file routing internally, which is more code not to fuss over.

It's also worth noting Nuxt consumed the fewest tokens of the bunch (3.85 million), though that did not make it the cheapest, given how Claude charges different rates for input and output tokens. Nuxt produced the app in 1,415 lines of code, which is second only to Hono/htmx's 1,370 lines.

Overall, these top three frameworks were within the margin of error of each other, which means any one of them could have gotten the top score depending on what kind of day the LLM was having.

Some of the more traditional frameworks lumbered way behind, however.

A SvelteKit-based app cost $4.28 to generate, with the agent repeatedly going back to the SvelteKit documentation to get a better handle on how the recently-updated Svelte works. Angular was the most expensive way for an agent to build an app, costing $4.97 while taking 17.2 minutes and chewing up 9.76 million tokens (with the agent failing to even finish a working app in two of its five attempts).

All the resulting apps aced the Lighthouse accessibility and performance tests. It's a pretty simple app to render, Lawson acknowledged.

Astro's app had the fastest loading time (called Largest Contentful Paint or LCP) of 810 ms, while the slowest frameworks in the cohort - Next.js (2,090 ms) and TanStack Start (2,028 ms) - took more than twice as long to deliver the goods, perhaps because of all the JavaScript they shipped to the client.

Lawson hopes this work will set the stage for applying a bit more rigor in selecting frameworks for AI jobs, now that developers won't have to do the actual heavy lifting themselves.

But a few developers on Hacker News pointed out AI efficiency isn't the only factor to consider. There's the human component of maintaining the framework as well.

"After a decade of React, I'm not going to go learn htmx or any other marginally better same-shaped tool for a new project, especially when it'd degrade my ability to review and debug another actor's work product," one wrote.

After all, the frameworks themselves might evolve with the help of AI, minimizing their differences over time. "Besides a rewrite to another framework using AI, another option is to modernize the framework with AI instead," someone else noted. ®

Twenty-nine NHS trusts listed as benefiting from Palantir's Federated Data Platform (FDP) said in Freedom of Information responses that they were not using it, according to health worker campaign group Medact.

That represents about 21 percent of the 137 trusts NHS England lists as "realising benefits" as ministers weigh whether to continue the contract.

The government must decide in December whether to continue the £330 million agreement with a Palantir-led consortium supplying the platform.

Medact, which is campaigning against the deal, collected the responses to scrutinize the government's case for retaining Palantir. The relationship between the US tech biz and the NHS began during the pandemic with contracts worth £60 million awarded without competition.

NHS England awarded the FDP contract in November 2023. It says the platform brings together operational data held in separate systems to help staff work more efficiently.

NHS England's September figures list 142 trusts as live on the platform - 137 of which it claims are benefiting.

Among the disputed entries, NHS England cites University Hospitals of North Midlands NHS Trust as receiving benefits from the platform, but in response to an FoI request in July, the trust said: "FDP is not live. We do not intend to roll it out in the future. A variety of factors prevent roll out, but primarily it does not appear to provide functionality over and above what we have already."

Meanwhile, a response from Nottingham University Hospitals said "no implementation timeline has been agreed," according to The Guardian.

The Register asked NHS England to explain the discrepancy.

An NHS spokesperson said: "The NHS Federated Data Platform is universally available for all NHS trusts and integrated care boards across England, with local NHS organisations able to select individual tools depending on their needs.

"137 NHS trusts have now reported benefits from the wide range of tools available on the platform - helping staff join-up and improve care for patients, and work more efficiently."

Palantir maintains its software is playing an important role in improving patient care and helping to deliver 100,000 additional operations, a 12 percent reduction in discharge delays and the removal of 675,000 patients from waiting lists.

In July, the UK's statistical regulator told NHS England to ensure greater accuracy in quoting figures following a row over data used to promote the purported benefits of the FDP.

Dame Chi Onwurah, the chair of the House of Commons Science and Technology Committee, has given the government until today to clarify its response to questions over whether NHS England has any exit plans should it decide to end the Palantir contract in February.

James Frith, a junior minister in the Department of Health and Social Care, wrote [PDF] to the Committee in October explaining that the contract has "measures built in to facilitate exit and transition including clear intellectual property rights in NHS build products or components."

He continued: "Publishing more detailed information on the Exit Plan would, or would be likely to, prejudice commercial interests and NHS England's ability to secure best value for taxpayers in future commercial activity."

Onwurah replied: "I understand that you will not publish the exit plan, but will you confirm if one has been prepared?"

The Register asked Palantir to comment. ®

Britain's AI infrastructure boom could soon have an unlikely new home: Drax Power Station, where the operator of the country's largest biomass power plant has unveiled plans for a £20 billion datacenter development capable of consuming more than a gigawatt of power.

Drax announced on Tuesday that it has teamed up with investment giant Brookfield and its UK-based AI infrastructure business Radiant to build the facility at its North Yorkshire site. The money would come through a joint venture between Brookfield and Qatar's Qai, although the £20 billion headline figure is more ambition than actual money on the table at this stage.

The companies plan to start relatively small, at least by the standards of the AI industry's apparently bottomless appetite for electricity. The first phase would deliver up to 100MW of capacity by early 2028, backed by a proposed £2.5 billion investment. A second phase could add another 500MW between 2028 and 2031, with further expansion taking the site beyond 1GW after that.

However, a few hurdles remain between today's announcement and a functioning AI megafactory. According to Drax's regulatory filing, the first phase still requires planning permission, regulatory and network approvals, commercial agreements, and final investment decisions.

It also needs a tenant, which is a fairly important detail when you're proposing to spend billions on somewhere to house other people's computers.

Drax reckons its existing infrastructure gives it an advantage over developers struggling to secure electricity connections. The 1,000-acre site already has access to 4GW of grid capacity, 2.6GW of active generation and cooling infrastructure, along with proximity to the UK's fiber network.

The UK government estimated last week that all of Britain's existing data centers use around 2GW of electricity. Drax's proposed development alone could eventually need more than 1GW, although how much power it actually consumes will depend on how fully it's used.

The first phase would draw electricity from the grid, but Drax wants to supply up to 500MW directly from its existing generating capacity under a long-term power purchase agreement, potentially allowing AI servers to run on electricity produced by burning imported wood pellets.

That brings another complication. Drax signed a government-backed Contract for Difference agreement last year covering biomass generation between April 2027 and March 2031. The agreement allows the company to request permission to redirect up to 500MW of capacity towards a datacenter, but ministers and the Low Carbon Contracts Company would have to agree, taking into account consumer value, energy security, and sustainability.

The announcement also arrives as enthusiasm for enormous AI server farms is proving rather less universal outside government and industry circles. Opposition to data center developments has been growing on both sides of the Atlantic, with communities raising concerns about electricity bills, water consumption, and the impact of industrial-scale facilities on their surroundings.

A recent investigation found that businesses accounted for 84 percent of the stakeholders UK ministers met to discuss datacenters between July 2024 and June 2026, while environmental groups secured precisely zero meetings.

Drax and its partners are promising a community benefits fund and say public consultation will begin this month. Whether that will reassure residents concerned about another enormous consumer of electricity is a moot point.

Drax CEO Will Gardiner said the project could "help secure the long-term future of the site and protect hundreds of jobs," while supporting Britain's ambitions to become a major player in AI.

The government, unsurprisingly, is enthusiastic. Chief Secretary to the Treasury Emma Reynolds welcomed the proposed investment as an opportunity to create jobs and support Britain's reindustrialization.

"Investments like this will help pave the way to a new age of British reindustrialization, creating good jobs in Yorkshire and delivering growth right across the country," she said.

Ministers may be enthusiastic, but Drax still has to find a customer, secure planning permission, and convince regulators before anyone starts plugging in servers. ®

Virgin Media O2 (VMO2) wants to chop a third of its labor costs and is expected to lay off thousands of workers, giving staff more to think about than Christmas presents during the run up to year-end.

Shrinking the workforce is one of a number of cost reduction measures designed to save the UK's second largest telecoms provider some £800 million by 2029.

It currently employs 16,000 people and redundancies will "equate to roughly 30 percent of the company's total people costs" over the next two years, VMO2 said on Tuesday.

The majority of the redundancies will come from "non-customer facing roles," it claimed. The Register asked the telco to be more specific on the headcount cuts and the type of roles are at risk.

A spokesperson refused to provide a number, saying those plans are being finalized and consultations are due to take place. Logically, if more senior head office positions are hit then fewer people will need to leave to meet the £300 million ($402 million) target figure.

The Reg understands thousands of roles are at risk.

Another lever management at VMO2 is pulling involves halting investment in any further fiber network upgrades beyond existing commitments. Instead, a low-cost upgrade will be deployed to its hybrid fiber-coaxial (HFC) network (i.e. cable customers) to bring 2 Gbps services to the majority of the network in 2027.

VMO2 was formed in 2021 by the merger of broadband and TV biz Virgin Media and mobile operator O2. The pair pledged to spend £10 billion (c $13.4 billion) on beefing up 5G coverage and laying gigabit "full-fiber" fixed-line networks.

Both Virgin Media and O2 talked up those connectivity commitments when justifying the case to combine their businesses, saying they needed greater economies of scale to compete with their rivals.

However, business has not gone as planned, with revenue falling over several successive quarters. The latest figures for Q3 ended September 30 show total revenue down 5.7 percent year-on-year to £2.43 billion. The networks giant blamed consumer fixed market pressures, among other things.

VM02 forecasts a revenue decline of 3 to 5 percent for the whole of 2026.

In the consumer division, VMO2 talked of more clearly defining the role for each of its brands. O2 Business is planning to streamline the product portfolio.

Tom Oughton, senior analyst at Megabuyte, said the changes were expected.

"VMO2 has undergone a strategic review as it looks to address rising leverage, with today confirming a transformation program to ultimately bring c. £800 million ($1bn) in cost savings ahead of its first debt expiry in 2029. The routes to achieve this are the most obvious ones: reduce capex on full fiber and focus on HFC upgrade (lower cost, quick payback) and reduce operational costs by cutting headcount," Oughton said.

"The obvious question is how a significant transformation will impact the business. VMO2 focuses on the benefits, but the business is declining 3-5 percent organically with 16-17k staff, so reducing headcount by 25 percent or more could stretch operations, worsen customer support, and cause broader disruption."

Paolo Pescatore, founder and analyst at PP Foresight didn't mince his words either: "Virgin Media O2 is in serious trouble, with customer losses and declining service revenue and earnings exposing the scale of the challenge."

"The business has lost its strategic direction, and continued customer departures suggest it is struggling to rebuild trust while rivals give people compelling reasons to switch. Cutting costs can buy time, but it cannot repair those relationships or compensate for weak execution," Pescatore added.

"Management must show customers why they should join and stay; otherwise, it risks shrinking the business without fixing what is driving people away. Shareholders need to take a hard look at these businesses and make even tougher choices about their future."

Noel McClean, deputy general secretary of Prospect union, said the job cuts "will be devastating for those directly affected and will of course have an impact on colleagues remaining in the business.

"Industry conditions are tough but this level of cuts will drastically alter the way VMO2 works." ®

The UK's National Cyber Security Centre (NCSC) says AI-powered malware is "highly likely" to be a part of a future ransomware attack on Britain's critical national infrastructure between now and 2028.

The warning accompanies its annual review and a separate assessment exploring how AI could change cyber threats over the next two years.

The cyber arm of signals intelligence agency GCHQ did not provide much detail about what form this AI malware may take. However, it alluded to open source AI security tools that were initially intended to assist security researchers being repurposed for nefarious activity.

Attackers are already repurposing such tools. For example, researchers are increasingly seeing widespread abuse of ARTEX, an open source, Chinese-made AI red-teaming platform, in attacks on UK registrar 123 Reg, Microsoft 365, and others since its launch in July.

In its pessimistic scenario for 2028, the NCSC envisages "a significant increase in the frequency and sophistication of cyberattacks worldwide," with wider adoption of AI tools making attribution harder.

China "has likely played a role" in supplying those tools in this future, "enabling them to hide their activities within the noise," the report states.

It also considers a further threat: an autonomous attacker operating beyond its human operator's control.

The NCSC says it "is a realistic possibility such an entity could self-replicate and conduct cyberattacks at will, with no apparent motivation or restraint."

Such attacks would likely complicate geopolitical relationships, owing again to attribution difficulties, making it easier for countries that order attacks on others to deny their involvement.

At a time when each Patch Tuesday seems to come loaded with record numbers of bugs, the NCSC acknowledges that AI is already changing vulnerability discovery and patching.

The body is also exploring how AI could help defenders keep pace. Project DROMOS began earlier this year with a four-week pilot analyzing government code repositories, before expanding into selected operational environments.

The Government Cyber Coordination Centre (GC3) launched the project with the AI Security Institute and other departments. The NCSC says it uncovered hundreds of potential security issues, including critical vulnerabilities.

The review also reaffirms plans for Cyber Shield, an agentic cyber defense system first teased in July.

Cyber Shield is expected to be delivered in stages over the next five years. It will involve the NCSC providing defensive agents deployed by participating organizations with data that will strengthen national resilience to ongoing cyber threats.

The NCSC will be running the scheme alongside the Department for Digital, Culture, Media and Sport (DCMS), but acknowledged that this kind of automated defensive activity currently sits outside many organizations' risk tolerances, meaning it has a role to play in building trust in the system before adoption can go mainstream.

As reported last week, the government has set aside £60 million to fund the initiative, and an additional £70 million to support a new National Exposure and Real-Time Vulnerability (NERV) Centre.

The NERV Centre's job will be to identify and maintain an ongoing understanding of vulnerabilities across CNI organizations and the wider economy.

Per last week's budget, £40 million will fund a National Agentic Security Unit responsible for responding to uncontrolled bots and other agentic mishaps, and £24 million was assigned to strengthen public sector defenses.

An undisclosed sum will also help establish a new NCSC-managed AI Cyber Security Centre in Manchester to develop AI security expertise across government.

In the meantime, NCSC CEO Dr Richard Horne is calling for organizations to "invest with urgency" in their security and resilience.

"For decades, we have collectively got away with doing tech on the cheap," he said today, announcing the annual report.

"We can be a nation where leaders of all organizations actively choose to invest with urgency... And where all organizations make security and resilience a fundamental requirement, not an afterthought.

"The collective result would be a source of enduring national strength for us all."

Horne said organizations should treat the NCSC's Cyber Essentials certification as the starting point, but scale up resilience efforts in line with organizational size, complexity, and risk.

The government also announced a 75 percent discount on Cyber Essentials certification for micro, small, and medium-sized businesses applying for their first certificate. ®

Apple says it sold more Macs in fiscal 2026 than in any previous year, but soaring memory prices are threatening to spoil the Cupertino-based gadget maker's record-breaking run.

The warning came during Apple's fiscal fourth-quarter earnings call on Monday, the first since John Ternus took over as CEO.

"What we're seeing with memory pricing right now is unprecedented," Ternus told analysts.

Apple has already raised prices across much of its hardware lineup, including Macs and iPads in June and iPhones in September. Those increases, along with existing inventories and cheaper components elsewhere, have softened the blow. Unfortunately for Apple, memory prices haven't finished climbing.

"For the December quarter, we expect even higher memory costs," Ternus said, adding that Apple expects to pay more for memory beyond the end of the year.

Mac buyers haven't been deterred so far. Apple pulled in $10.8 billion from Macs in the September quarter, up 24 percent year-on-year, following $10.35 billion in the June quarter, up nearly 29 percent year-on-year. The MacBook Neo and MacBook Pro accounted for much of the latest growth.

"We have never sold more Macs in a fiscal year than we did this time," Ternus said.

Whether Apple can keep enough of them on shelves is another matter. COO Sabih Khan said the company had already experienced supply constraints across iPhone, Mac, and iPad during the September quarter, largely because of limited availability of advanced chipmaking capacity. Now memory is taking over as the chief headache.

"For the December quarter, we expect constraints on Mac and iPad, primarily driven by memory, as demand has been higher than what we expected," Khan said.

Apple isn't sitting idle. Ternus revealed that its manufacturing purchase obligations jumped by $33 billion between June and September as it bids to secure supplies, although he offered no indication of when the memory crunch might ease.

Higher memory costs are also squeezing Apple's margins. CFO Kevan Parekh said gross margin fell by roughly one percentage point between the June and September quarters, excluding tariff effects, entirely because of memory costs. The damage would have been worse without existing inventory, cheaper non-memory components, and those price hikes.

Apple expects its gross margin to fall from 47.9 percent in the September quarter to between 45 and 46 percent in December, primarily because of higher memory costs. Costs associated with rolling out Siri AI will also weigh on the figure.

Parekh said Apple had "reluctantly" raised hardware prices, but supply constraints made it too early to assess how customers were responding.

Meanwhile, Ternus is keen to sell customers on the Mac as an AI machine, particularly for running models locally.

"Mac mini and Mac Studio in particular are pushing the boundaries of local private AI," he said, pointing to workloads that previously would have been difficult to run on desktop hardware.

Apple has also refreshed its Mac lineup with M6-powered machines and a touchscreen MacBook Pro, giving customers more reasons to upgrade, provided there's enough memory to build the things.

Not that the wider business is struggling. Apple reported September-quarter revenue of $114.5 billion, up 12 percent year-on-year, while iPhone revenue climbed 17 percent to $57.6 billion. Services brought in $31.6 billion, up 10 percent.

Apple forecasts revenue growth of 7 to 10 percent in the December quarter despite the supply constraints. Even with memory prices running wild, it isn't expecting to go hungry. ®

Euro cloud trade association CISPE can keep the identities of members backing its VMware merger challenge confidential from Broadcom following a ruling by the EU General Court.

Cloud Infrastructure Services Providers in Europe filed an action last year seeking to annul the European Commission's approval of Broadcom's acquisition of the cloud and virtualization giant.

CISPE believes the Commission erred in allowing the purchase to go ahead without imposing any conditions on Broadcom to protect the many European cloud providers and their customers who depend on VMware's software.

Broadcom sought access to redacted information identifying members supporting the action, arguing that it needed the information to challenge the action's admissibility. CISPE says the ruling allows the case to proceed without disclosing those identities to Broadcom.

"It is a procedural ruling on confidentiality rather than a decision on the substance of the case, but it removes the immediate obstacle to the action moving forward," a CISPE spokesperson told The Register.

CISPE argued that identifying the members, which are or have been VMware customers, would expose them to a significant risk of retaliation because of Broadcom's market power. Those members had supplied information about their commercial relationships with VMware.

As far as the case itself goes, CISPE says it warned the Commission that Broadcom would use VMware's pre-eminent position in the private cloud market to rewrite licensing terms and reshape the market in its favor.

Just a month after Broadcom's acquisition concluded in November 2023, partners were informed that VMware's channel program was to end in early 2024, and that Broadcom was also ditching the majority of VMware's Cloud Services Providers, keeping only a select few that had the largest turnover.

Despite Broadcom CEO Hock Tan publicly stating that inflating prices was not his strategy, that is what happened. AT&T says that Broadcom proposed a 1,050 percent price increase, for example. Meanwhile, nine out of ten respondents to a recent survey of VMware customers said they were considering moving to alternatives.

CISPE, which represents 48 cloud infrastructure service providers across the EU, says fewer than 20 European providers remain authorized to offer VMware-based services, down from more than a thousand.

Commenting on the latest ruling, CISPE Secretary General Francisco Mingorance declared it a good day for the organization and for the rule of law in Europe.

"The case matters even more today than when we filed it. In the three years since the European Commission waved this deal through without conditions, Broadcom has dismantled the European VMware channel," he said.

"Those excluded have until 31 March, 2027, to sell their infrastructure or their customers to a competitor chosen by Broadcom, or to close. This is the foreclosure the European Commission was warned about."

We asked CISPE what outcome it hoped to achieve, nearly three years after the acquisition closed.

"The immediate objective is to have the Commission's unconditional clearance annulled, on the basis that it failed to properly assess the effect of the acquisition on European cloud providers and the VMware channel," CISPE's spokesperson told The Register.

The Commission did impose conditions concerning interoperability with rival hardware, but CISPE argues these did not address the risks to cloud providers.

"We appreciate that it's a long shot (although there is precedent)," the spokesperson continued. "We have (along with others) also filed a competition complaint with the Directorate-General for Competition against Broadcom for its actions post-merger."

The precedent referred to is the EC approval given to the merger of Sony and BMG's music divisions in 2004, which was ruled to be not of "the requisite legal standard" two years later. The deal was subsequently reassessed and re-cleared.

"Ultimately, we want to do whatever we can to bring Broadcom to the table to negotiate and ideally end some of its most egregious practices so that European cloud providers can once again support cloud customers in a commercially fair and viable manner," the spokesperson added.

The Register asked both the European Commission and Broadcom to comment, but neither had responded by the time of publication. ®

Public bodies in the UK expect higher operating costs and energy charges as a result of climate changes and the transition to lower carbon energy, but few have a good picture of the risks, according to government auditors.

A new report from the National Audit Office (NAO) says some public bodies expect financial impact from the risks they identified, although bodies were often not able to cost these - even approximately.

"Bodies anticipate costs to arise from damage to assets and infrastructure, driven by physical damage from extreme weather, as well as from chronic shifts to higher temperatures and more precipitation. Bodies also expect higher maintenance and repair costs, potential asset upgrades, and transition costs. Many bodies expect increased operating costs, including energy charges. Bodies controlling significant property or infrastructure are particularly concerned with decarbonization and adaptation costs," the report stated.

His Majesty's Treasury (HMT) has set out rules for climate-related financial disclosure in larger central government bodies' annual reports, which produced their first set of reports for the 2025-26 financial year. Other rules for disclosure have been in place since 2022.

However, the NAO found [PDF] that understanding of the risks in the impact of climate change - increasing temperatures, higher risk of flooding, higher winds and longer droughts - was poor, as was insight into costs of moving to a low-carbon economy.

"Bodies did not articulate transition risks as well as physical risks in the strategy disclosures. In the disclosures we reviewed, while several transition risks were listed, few specifics were given, unlike physical risks, which were explained in greater detail," the report from the watchdog stated.

"The transition risks most frequently identified included the financial cost of transition, and supply chain disruption. Anticipated transition costs included the costs of: switching to lower-carbon energy sources, including investments in new technology; complying with new regulations; and of fluctuations in the price and availability of materials, water, energy and specialist skills," it added.

Gareth Davies, head of the NAO, said: "UK public services will need to adapt to the impacts of climate change. Parliament needs clear information on the resilience of infrastructure, service provision and finances to climate change, and on how these risks are being managed."

"Although public bodies are still learning how best to report their most pressing climate risks, the requirement on all large central government bodies to make such disclosures from 2025-26 is an important first step. We expect this reporting to become more useful over time."

Earlier this year a report from Global Action Plan, a climate charity, claimed 84 percent of proposed datacenter developments were in areas that are already water stressed or projected to be water stressed by 2040.

Water consumption by datacenters could intensify strain on the UK's water systems already stressed by climate change, it argued.®

Raspberry Pi has announced yet another round of price hikes, this time adding $80 to the cost of its 16GB Pi 5, which now sells for more than three times its original launch price.

The latest bump, announced by Raspberry Pi boss Eben Upton, takes the 16GB Pi 5 to a whopping $385. The 16GB Compute Module 5 variants are also getting $40 more expensive, although the Raspberry Pi 500+, which packs the same amount of RAM, has escaped this particular round.

Upton blamed a familiar culprit: soaring memory costs, worsened by a lack of alternative suppliers for 16GB DRAM and ongoing shortages of other components.

"Unfortunately, a lack of alternative supply for 16GB DRAM, compounded by other ongoing component shortages, means we now need to make further price increases on some of our highest-density products," Upton wrote.

It's becoming an expensive habit for Raspberry Pi customers. The 16GB Pi 5 cost $120 when it launched in January 2025, which seems almost reasonable now. Raspberry Pi added $25 in December, another $60 in February, and a further $100 in April. With today's $80 increase, buyers are being asked to cough up $265 more than they were less than two years ago, a price hike of 221 percent.

And it's barely been a month since Raspberry Pi raised prices on its 2GB Pi 4 and Pi 5 models by $12.50 apiece, taking them to $67.50 and $77.50 respectively.

The repeated markups come despite Raspberry Pi reporting record first-half results in September. Revenue jumped 90 percent to $256.9 million, while gross profit rose 79 percent to $59.4 million. Gross profit per board climbed 53 percent to $12.20, helped partly by memory purchased before prices went through the roof. That stockpile advantage has since largely disappeared.

For customers unwilling to pay nearly $400 for a single-board computer, Upton advises buying something with less RAM, using an older Pi, or optimizing applications to make do with less memory.

He also warned that some sellers are exploiting the shortages by charging inflated prices or offering products that aren't what they appear to be, urging customers to stick to approved resellers.

"We're aware that a few bad actors have been taking advantage of the current memory supply circumstances to sell Raspberry Pi products at inflated prices or shift products that aren't what they seem," he said.

There is, apparently, some good news, as Raspberry Pi continues to insist the increases are temporary. "As we've said from the beginning, the current memory landscape is difficult but ultimately temporary, and we look forward to rolling back our price increases when memory costs fall once again," Upton wrote.

That could be a long wait. Micron has previously warned that memory shortages could drag on into 2028 or beyond, so anyone hoping to see a $120 Pi 5 again probably shouldn't hold their breath. ®

Distributed teams need to collaborate on shared information, but duplicated files, delays, and competing versions keep getting in the way.

Picture a video production team with editors in different cities. One person updates a project file, another keeps working from an earlier copy, and a third waits for a large transfer to finish.

When teams need to collaborate on large files across locations, traditional storage and file-sharing arrangements introduce friction precisely where people expect technology to help.

The hard part isn't merely moving a file. It's knowing who has the authoritative copy, when changes are available, and whether somebody else is still working on an outdated version.

Consider a game studio preparing a build for review. A designer exports an updated asset, but a colleague pulls an older version from a synced folder. The mistake only becomes obvious when the pieces are assembled. The team must trace the mismatch, resend files, and repeat work.

Sync-and-download workflows depend on a chain of uploads, downloads, and decisions about which copy is current. Under deadline pressure, that chain breaks. Delayed access means missed handoffs; duplicated files mean extra checking and sometimes rework.

For IT, each workaround creates another question: where does the data live, who can access it, and how do teams maintain consistent controls? Adding another collaboration application doesn't resolve those underlying questions.

The useful starting point is the workflow. Find out where teams copy data, where they wait, and what prompts them to work outside the shared system.

These frustrations cut across industries, but the ways organizations respond aren't identical. Hearing how another team handled version confusion, access, and handoffs is often more useful than another abstract promise of seamless collaboration.

That's the focus of The Register and LucidLink's invitation-only dinner on November 24 at Covent Garden Hotel in London. Joe Fay of The Register will moderate a peer discussion under the Chatham House Rule.

Bring the real situations that never fit neatly into a technology diagram, and hear how others are handling them.

Request your place at the November 24 dinner

Sponsored by LucidLink

The outputs of AI models are the result of a complex interplay between neurons, weights, training, and inference, and are therefore difficult to follow.

But in some ways they're starting to look more like open books. Recent work by Anthropic safety fellow Joshua Fonseca Rivera has revealed how certain neurons in a neural network can be associated with a specific persona.

Now, computer scientists from MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) have found that frontier AI models like OpenAI's GPT-6 Astra can rewrite the weights of another model directly, as opposed to indirectly through training, and alter their behavior.

"We find early signs that frontier LLMs are surprisingly good at natively reading, writing, and reasoning about neural network weights," said Rikiya Takehi, a PhD student at MIT CSAIL, in a social media post. "Instead of training, models may treat raw weights like readable, editable code: diagnosing failures and patching them directly."

To demonstrate that, Takehi and colleagues Jayden Teoh, Alex Zhang, and Omar Khattab asked GPT-6 Astra to edit Qwen3.5-0.8B weights to make it better at math.

Astra changed only 0.016 percent of the model's weights, reducing its reasoning loops and improving its math performance by as much as sixfold on the MATH-500 benchmark, while boosting its instruction-following score on the IFEval benchmark roughly threefold.

There was no training using gradient descent; Astra just revised the numerical values that constitute Qwen's weights.

"All of this is possible without allowing Astra to write training loops," the researchers wrote in a blog post. "It uses its reasoning to figure out what each part of a neural network computes, or should compute, and changes the weights accordingly. We call this Weight-Space Programming.

"Throughout the history of deep learning, we encoded our simplistic inductive biases into tiny symbolic optimizers like Adam or Muon. This is changing: our neural network weights will soon be natively optimized by LLMs that deliberately reason about their weights."

Astra's ability to program weights extends to being able to write a neural network from scratch, zero-shot. The researchers told Astra: "write, from scratch, the weights of a policy network that plays [Breakout] as well as possible." And it did so.

Takehi suggests this technique shows promise for improving models without the consequences of catastrophic forgetting, which can occur when a model is revised to include new knowledge.

More generally, the authors expect Weight-Space Programming will open the door to new ways to understand models and optimize them, and to implement recursive self-improvement. The technique may also prove useful for reducing model sycophancy and improving interpretability and alignment.

The next step, the authors say, involves building a benchmark and a harness for carrying out Weight-Space Programming on LLMs. "We believe this will provide a standardized evaluation to measure and accelerate progress in Weight-Space Programming, turning what currently seems like black magic into a rigorous, reproducible engineering discipline," they say. ®

For-profit model-makers, many of which are spending billions on AI infrastructure, will be chasing revenue from just ten percent of steps involved in agentic AI workflows, according to analyst firm Gartner.

"Nine in 10 agent workflow steps are routine, and open-weight models handle them at parity for a quarter of the price of the cheapest closed model," the firm's senior director Mark Petty and group vice president Raymond Paquet wrote last week in new document titled "Open-Weight Models Will Run 90% of Enterprise Agentic Workflow Steps."

"Typical operational workflows involve hundreds of routine executions, such as classification, tool calling, passing, validating and reconciling data," the pair observe, before suggesting open-weight models "can handle hundreds of these routine steps for every one that needs frontier reasoning."

"Against the cheapest closed model at the same score, the open model costs a quarter as much. Against a frontier model run on every step, it costs as little as a twentieth," the pair add.

Those are the best-case numbers, because the analysts admit "Current open models can be more verbose and retry more often, so the saving per task against a frontier model is closer to five or six times."

That's still a lovely saving, but the analysts rate it as the welcome by-product of moving to open models rather than an end in itself.

"Years of workflow history are a large part of what an acquirer pays for in a SaaS company and every routine step sent to a frontier API moves a slice of it outside your perimeter," the pair wrote. "Open-weights preserve your control. The cost saving is a bonus."

The two analysts therefore offer the following strategic planning assumption: "By 2030, 90% of enterprise operational agent workflow steps will execute on open-weight models."

They argue that enterprises will mostly adopt domain-specific open-weight models, because they do well "in scenarios where a small number of highly stable steps run millions of times a month."

Most models, however, are already commodities. Harnesses, the tools that run agents and manage agents, are the source of value in AI products, because one of their roles is to route work to the most appropriate model for a job, without users noticing any impact. A good harness saves money, imposes strong guardrails, and manages agent fleets. Less powerful harnesses leave users with more to do.

The paper suggests frontier models will continue to have a role as the engine of research and development and are effectively capital expense. Open-weight models become an operational expense after frontier models do the heavy lifting. ®

A cloud architect has created an open-source alternative to VMware's Virtual Disk Development Kit (VDDK), which the virtualization giant recently withdrew from public downloads.

VMware describes VDDK as "a set of libraries and utilities that enable software developers to create applications that can access and manipulate virtual disks" and "a valuable tool for developers of storage and backup applications."

Users, along with VMware's rivals, had another use for VDDK: migrating VMs to alternative virtualization platforms.

In August, VMware restricted access to VDDK downloads to its partners, meaning VMware customers couldn't access the tool and some of its rivals' migration tools became less effective.

Cloud architect Andrija Panić decided to do something about it.

"Many tools that have nothing in common except VMware use the same small library to read a virtual machine's disks: virt-v2v and nbdkit, the importers in Forklift/MTV, Proxmox VE and Apache CloudStack, and countless in-house scripts," he told The Register by email. "When public VDDK downloads ended, all of them became harder to install and run."

"I didn't think it was fair that people's ability to move or protect their own data should depend on a download page."

He describes the project as "an independent drop-in for libvixDiskLib.so.8.0.3, the disk library that VMware backup and migration tools load to read virtual disks. It was written from scratch, using only public documentation and black-box testing, and contains no VMware code."

Panić told The Register he wrote the tool in Go, with a thin C layer that exposes exactly the same C interface, library name and symbol versions as VMware's code.

The tool cannot write disks, just read them for backup and migration purposes.

In his email to The Register, Panić said he has successfully tested OpenVDDK on vSphere versions 6.7 U3, 7.0.3, 8.0.3, 9.1.1. He said that tests in his lab with the latter version, accessing OpenVDDK through vCenter on a standalone ESXi host, saw every disk read match.

"Independent testers have reported successful use with Forklift (VMware to KubeVirt, cold and warm), MTV on OpenShift (cold and warm), a Proxmox VE pipeline and custom nbdkit-based tools," the architect told The Register. "I've received reports of production use, including a German cloud provider migrating a large file server with V2K Migrate and a US-based company migrating to Apache CloudStack/KVM."

Panić was at pains to point out that the reports of successful use are "not independently audited customer endorsements or a guarantee of suitability for every production environment." As is often the case with FOSS tools.

OpenVDDK is not the only option for migrating away from VMware without VDDK. KVM-based OpenNebula has a tool called OneSwap that it says has a 96 percent success rate. Nutanix recently tweaked its "Move" migration tool so it can do the job without VDDK. Red Hat is working on a fix but has provided details of a workaround.

Panić made OpenVDDK available under an Apache license, so perhaps Red Hat and others will find it useful as they re-jig their VM migration tools. ®

Japanese carmaker -maker Toyota has decided to have a crack at the market for AI processors.

Toyota is the major shareholder in a company called AISIN, a manufacturer of automotive components. Both companies are aware that cars are increasingly reliant on tech like image recognition, and that robotics and other forms of "physical AI" will need parts that overlap with cars.

AISIN thinks existing chips aren't a good fit for such roles, because they rely on external memory to store AI models and then generate a lot of traffic between a processor and the model. That traffic, the company contends, makes image recognition and other on-device uses of AI less effective. DRAM is also currently vastly expensive.

The company's response is a system-on-chip built on a 16nm FinFET process, an Arm Cortex A53 dual-core CPU, plus an Arm MaliG31 GPU. The chip includes built-in STT-MRAM, which stores both model weights and an operating system.

AISIN and its research partners at Tohoku University say that approach means the chip can go from power-on to kernel startup in 65.7 milliseconds - 38 times faster than is possible if using external memory - and run the Tiny YOLOv3 object detection algorithm on the chip.

"To effectively utilize the limited on-chip memory capacity, a hybrid multiplication architecture combining 8-bit and 16-bit multiplication was developed" according to a machine translation of Tohoku University's announcement [PDF]. "By combining 16x16, 16x8, and 8x8 bit multiplication, the amount of weight data for the AI model was reduced by 53 percent, from 17.7MB to 9.36MB, while maintaining 16-bit precision."

AISIN is an established parts supplier and already sells to Toyota and other automakers but hasn't said if or when it can get this chip into production.

Meanwhile, the likes of Nvidia and Qualcomm very much want to sell their silicon to carmakers, and position themselves as sources of tech and AI smarts that existing automotive supply chain participants can't match. ®

Anthropic has published a regulatory filing revealing nearly $900 billion in compute and infrastructure commitments, with additional estimated payments potentially pushing the total beyond $1 trillion.

The Monday filing reports that Anthropic expects to pay nearly $70 billion in contractual obligations by September 2027, with a total of roughly $447.5 billion in commitments expected to be incurred by the end of 2030. That's assuming of course its revenues continue their upward trajectory, which the prospectus warns is far from guaranteed.

"In the nine months ended September 30, 2026, our revenue was $33.2 billion, representing a year-over-year growth rate of 1,206%," the prospectus reads. "We do not expect our revenue will continue to grow at comparable rates, and we cannot assure you that it will continue to grow at all."

"Our ability to meet the substantial and growing financial obligations of these requirements to secure compute depends on our ability to generate significant revenue growth," the filing reads.

Which is to say, if Anthropic's revenues don't grow faster than its compute commitments, it could end up falling into a nasty spiral. And while Anthropic has several options, including debt and equity financing, relying on these long term could dilute existing shareholders or force the company to borrow on unfavorable terms.

Every IPO filing reads a bit like this because no company can guarantee eternal success and growth. A lot of this filing's language, as damning as it might sound, is therefore just Anthropic covering its butt.

However, even though the company is still growing steadily year-over-year, the latest filing shows revenue growth is already slowing. Between Q1 and Q2 of this year, revenues grew nearly 2.5x. Yet in Q3, Anthropic's reported revenues of nearly $17 billion represented an increase of just 46 percent from Q2. In the same period, Anthropic's compute and infrastructure costs grew 77 percent and it kept signing deals that commit it to more expenditure.

Last week, we reported that Anthropic's purchase commitments exceeded $800 billion over the next decade or so.

Anthropic now says its commitments are closer to $895 billion.

Roughly $320 billion of those commitments relate to existing contractual obligations. Another $451 billion covers leases Anthropic has signed, but that have yet to commence, with the company expecting them to begin between 2026 and 2028, although some remain subject to termination rights or other conditions.

Overall, however, Anthropic's compute spend is on track to increase substantially over the next few years. And that's not even counting the blank checks Anthropic has written to secure adequate compute, which it values at $150-$175 billion, putting Anthropic's total compute tab at somewhere north of $1 trillion.

Anthropic can't walk away from some of this spending because its deals with Microsoft, Google, Amazon, and Broadcom, which now total $437 billion, are non-cancellable. Some require payment regardless of whether Anthropic lights up a single GPU.

The AI upstart estimates that of its $895 billion in contracts, only about $213 billion worth allow it to avoid payments.

The revised S1 filing also offered insights into the company's customer base, including a major reliance on cloud service providers for distribution. The likes of Amazon, Google, and Microsoft are some of the model dev's biggest compute providers, but according to the filing, cloud marketplaces accounted for roughly 44 percent of its revenue in the nine months ended September 30.

Anthropic insists that none of its customers account for more than 10 percent of revenue. Yet reliance on hyperscalers for distribution remains a risk because they are all either actively developing competing products or provide access to models from Anthropic's rivals.

Despite all the risks in its prospectus, Anthropic believes it can become a sustainable business if it can access enough compute capacity. And by doing so, the prospectus claims AI will reshape the economy "in ways greater than the impact of industrialization, electricity, semiconductors, and the internet."

With so much at stake, how can humanity afford not to spend like a drunken sailor? ®

Unknown criminals are exploiting a maximum-severity vulnerability in Node-RED, a flow-based programming tool that is widely used in critical infrastructure, to deploy a flow that downloads and runs a Linux bot.

Node-RED is commonly used for workflow automation in industrial control systems (ICS) and operational technology (OT) environments. It's built on top of Node.js and is used in commercial products including Siemens SIMATIC IoT2050. In August, Siemens assigned the vulnerability identifier - CVE-2026-58115 - to describe the bug and explain its perfect 10-out-of-10 CVSS rating.

Affected Siemens devices do not enforce authentication on the Node-RED HTTP interface, allowing unauthenticated access to programming nodes that can execute system commands on the device. This means unauthenticated remote attackers can create malicious flows, which is what VulnCheck researchers began documenting last Friday.

"On October 30, 2026, one of our canaries caught an attacker exploiting CVE-2026-58115," VulnCheck researcher Patrick Garrity warned in a Monday advisory, adding that two Node-RED built-in nodes make this vulnerability "trivial to weaponize."

One is an inject node, which fires a message either on a schedule or once at startup. The second, an exec node, can run shell commands. "Wire one to the other and the flow executes a command as soon as it is deployed, with no further interaction," Garrity said.

Plus, these exploits come as criminals increasingly look to target internet-exposed OT and ICS devices, allowing their cyberattacks to impact the physical world and produce real-life harm. Case in point: the ongoing refrigeration controller hacks across Europe and the attacks on water systems across at least 12 US states over the summer.

The activity observed by VulnCheck came from a single IP address, 4.201.105[.]116, which sits in Microsoft's AS8075 network and geolocates to Brazil. "It didn't just scan us. It sent the complete exploit, flow and dropper, in a single POST /flows," Garrity wrote.

While there's a lot that the researchers don't yet know about the attacks - they didn't retrieve a binary or staged scripts to analyze - the filename, bot-linux-amd64, indicates that the malware targets Linux running on AMD64 processors. It also suggests the malware operator may have other builds. "We did not fetch the binary, so we can't say what it does once running," Garrity noted.

Between October 30 and November 1, the researchers saw this IP address target the same Node-RED vulnerability 16 times, and log 125 hits against eight different CVEs across their wider canary network. Among the targets of the attack were a seven-year-old Apache Solr flaw tracked as CVE-2019-17558 - affecting Camunda, Gitea, MLflow, Traefik, and Joomla, alongside Node-RED. All this suggests "opportunistic mass scanning rather than a targeted attack," according to the researcher.

One of these hits delivered a different payload pointing to a separate server (85.215.73[.]250), which VulnCheck now tracks - but doesn't have any telemetry other than the URL. "Because the download host is a different machine from the one that sent the exploit, the likely setup is a scan-and-infect arrangement: the hosts that find and exploit vulnerable instances pull the implant from a separate, dedicated server," Garrity wrote.

The mix of old and new bugs across unrelated products indicates "opportunistic mass scanning rather than a targeted attack," Garrity wrote.

"If you run Node-RED anywhere, including inside OT environments, check today whether it's reachable from outside your network, apply updates where available and implement detections to reduce the risk of exploitation," he added in a LinkedIn post.

While the bug has been added to VulnCheck's Known Exploited Vulnerabilities (KEV) catalog, it hasn't appeared (as of press time) in the US Cybersecurity and Infrastructure Security Agency's KEV.®

The next time you see a software engineer shaping silent words with their mouth, you may have to think twice before asking if they need a mental health day - they might just be wearing a pair of smart glasses that let them talk to their computer without uttering a sound.

Subvocal is a Y Combinator-backed startup aiming to produce these glasses, and already has working prototypes to show off in YouTube videos, including one with YC partner Dave Lieb trying them out and being impressed enough to offer funding. The headset isn't what the final version will look like, the company notes, which is good: The demo headset is obviously 3D printed and far from fashionable.

Once the final product is available it should look a lot sleeker, based on mockups on the company's website. It's not clear when they'll actually be released, though the preorder page says it'll be sometime in Q1 2027 at an early bird price of $499. It's not clear what the final full price will be.

As for how the glasses work, the founders, Georgia Tech computer science students Aarush Yadav and Arnav Kumar, explain in a launch post on YC that the device uses low-powered radio waves projected from the glasses toward a wearer's face. Those waves are distorted by muscle movements and the change in a person's face shape as they speak or simply mouth words, which the duo said can be picked up and translated into written text.

"We started working on this about four months ago after building the first version at a hackathon," the pair explained in their YC post. "Since then, we've gone from a table full of electronics and antennas taped to our face to a semi-portable board and sleek glasses that can decode open-ended speech at 200+ WPM and a reasonably low error rate even at a small dataset scale."

This is just a proof of concept, and many key details are still missing, like the accuracy rate, battery life, whether it will paired with an app, or what-have-you.

"The core thing we wanted to prove first was that the signal exists and that we can decode it," Yadav and Kumar said.

In a Hacker News post announcing the company, Yadav and Kumar explained that they started the project during the hackathon by focusing on using EMG electrodes attached to Yadav's face. They didn't win the hackathon, but apparently the demo was impressive enough to earn some initial funding from San Francisco VC firm Afore Capital.

Shortly after that, the pair pivoted from EMG to RF sensing, as a commercial dictation product that required sticking electrodes to one's face was unlikely to succeed.

"Despite prior academic research, getting this to work was a huge engineering challenge," Yadav said. "First we built an extremely low-noise RF sensing stack. Then we had to place antennas so their fields overlapped maximally with the articulators. Our first attempt on glasses failed completely, so we reverted to old habits and taped antennas on our face."

They eventually got to the functional prototype they have now, and pre-trained the AI models doing the speech recognition to minimize setup time. According to the company, Lieb only had to read 20 sentences to calibrate the headset to his speech patterns.

"We think silent speech can become the next interface for interacting with computers, replacing voice and the keyboard," the company says of its goal. "Emails, prompts, messages, code, notes, commands, all at the speed of speech without bothering anyone around you."

With just four months of work under their belts, Subvocal could be a massive hit, ushering in a new era of silent muttering productivity for workers in a variety of industries, or it could flop as it tries to scale.

Those interested in finding out can preorder a pair now for $199, with the other $300 not due until the glasses ship. ®

If AI can find the holes in programming code, it can do the same with legal code, paving the way for chaos.

Felix Ringe, a visiting fellow at Stanford Law School's liftlab and legal scholar, suggests large language models (LLMs) have become potent enough that they could break legal systems by identifying exploitable inconsistencies at scale.

"Legal systems encode the foundations of social order in natural language," Ringe and collaborator Linus Folkerts argued in a blog post last week. "Similar to computer systems, the written law can contain vulnerabilities that invite exploitation."

Ringe contends that while courts tend to pave over these text vulnerabilities through statutory interpretation, the ability to discover adversarially useful inconsistencies in the law in an automated way risks overwhelming the legal system's ability to bridge the gaps.

To buttress his thesis, he points to recent work by other legal scholars that frames laws as vulnerable code, such as "Legal Zero-Days: A Novel Risk Vector for Advanced AI Systems" and "Hardware and Software Bugs in the Legal System" [PDF].

The analogy dates back at least to 1999, with Lawrence Lessig's Code and Other Laws of Cyberspace [PDF] and Code Is Law: On Liberty in Cyberspace. Lessig credits William J. Mitchell, in City of Bits: Space, Place, and the Infobahn (1995), with the idea that code is cyberspace's law, adding that he "drew the metaphor from others as well."

Ringe has turned this on its head by proposing that law is code.

The analogy between programming code and legal code works reasonably well, up to a point. Law and code both have their share of formatting pedantry and ritual. The mapping breaks down when courts get cast as a runtime environment - non-deterministic trial outcomes don't work for production applications.

Discontinuities aside, Ringe's point is that AI models can surface statutory gaps at a scale that may bring the chaos currently seen in the software security ecosystem to the legal world.

To demonstrate that, Ringe and Folkerts cite several potential gaps in German law that they uncovered with the help of LLMs, noting that they're only sharing legal exploits they believe could not cause serious harm.

For example, an amendment to Germany's Occupational Diseases Regulation (BKV) is subject to a constitutional requirement to cite its statutory authority. But Ringe and Folkerts argue that the amendment omits a required reference. So, in theory, a journalist could request an unpublished statement submitted to Germany's occupational-diseases advisory council under freedom of information rules, challenging a confidentiality provision that the researchers argue may be invalid.

Another example cited describes a transit visa rule that may be invalid because the government failed to identify the authority under which the rule was made.

"We believe that with cheap legal reasoning proliferating, AI systems could jeopardize the functioning of today's legal systems," they wrote. "In such a world, lawmakers would need to significantly harden the process of legislation, for example by leveraging advanced AI themselves."

Wendy Musell, a San Francisco-based employment law attorney and partner, told The Register that the use of AI in the legal field is expanding rapidly.

"Some California Courts, including LA Superior Court, are testing the use of AI as law clerks," said Musell. "Generally law clerks would assist the Court by researching and drafting opinions for the judge's review. That means AI is serving that role in certain courts. Litigators now must consider how AI is analyzing motions and cases.

"This is the case even where I personally have found different AI sources to appear to have a slant or bias, and often fail to accurately analyze legal issues. It was just a matter of time before AI came for the legislative process."

Eric Goldman, associate dean for research and professor of law at Santa Clara University School of Law, told The Register that Ringe's argument has some merit.

"I think the methodology of using AI to identify potential statutory gaps is logical," said Goldman. "However, it will require careful human review to determine if the AI-identified gaps are in fact actually gaps. Statutory gaps also may have no legal consequence, or at least none of significance; and legislatures will often correct any major gaps once they are identified."

Goldman said the idea that legislators should use AI during the drafting process is ambitious, at least in the short term.

"Legislators often draft laws with a range of goals and priorities, and avoiding statutory gaps is rarely a high priority," he explained. "But many legislatures have legislative counsel who are responsible for converting draft bills into final legislative text, and I could see where legislative counsel could benefit from using AI to identify drafting gaps as part of their standard cleanup protocols." ®

A coalition of private-sector security firms working with Japan's National Police Agency has disrupted XCPE, a backdoor used to build a massive botnet comprising more than 1.3 million Android devices across more than 200 countries.

Google, Lumen, Human Security, Plume, Synthient, and TrendAI participated in the disruption.

"We assess with high confidence that the backdoor and botnet were intentionally developed, deployed, and operated for financial gain by Videostrong Technology," the Google Threat Intelligence Group (GTIG) said in a Monday blog.

Videostrong Technology is a hardware manufacturer based in Shenzhen, China. It builds Android devices, including home TV boxes and projectors. According to Google, Videostrong deliberately embedded the XCPE backdoor in device firmware and sold access to criminal affiliates, allowing them to install additional malware modules on devices shipped to unsuspecting customers. When buyers connected these devices to their home networks, the preinstalled backdoor allowed operators to create proxy networks and hide malicious traffic using the customers' internet connections without their knowledge.

The firmware's pre-installed management agent purported to be a standard remote-management client, according to TrendAI. It pulled updates from a GitHub page, and connected to a Message Queuing Telemetry Transport (MQTT) server that gave the operator full remote control of each box, allowing them to install and remove apps, change settings, and push updates to the infected devices.

Between June 3 and October 1, TrendAI documented more than 3.8 million distinct home and mobile IP addresses connected to the operation.

The coalition also disrupted the command-and-control domains used to manage the affiliate modules. "Because these affiliates each have diversified distribution channels for their modules beyond XCPE, we assess that taking down their C2 domains will sever access to at least 10 million additional compromised devices," according to GTIG.

Google said it had since cut ties with Videostrong, and banned the OEM from making devices with preloaded Google apps.

Videostrong did not immediately respond to The Register's inquiries.

This backdoor and botnet operation is part of the same ecosystem as BadBox 2.0, another notorious botnet that infected more than 10 million TV boxes and other devices before Human, Google, and TrendAI disrupted it in March 2025.®

Hoping to dominate the next phase of AI networking, Marvell Technology plans to acquire photonics giant Lumentum in an all-stock deal valued at $115 billion. The move will fill several key gaps in Marvell's photonics portfolio, particularly with regard to high-performance laser sources, optical transceivers, optical circuit switches, and specialized compound semiconductor manufacturing capabilities involving indium phosphide.

Higher speed interconnects combined with larger compute domains are the driving forces behind the move to photonics. As interconnects push beyond 400 Gbps per lane, copper interconnects are being pushed to their limit, necessitating denser, hotter rack systems. Optics offer a reprieve from these scaling laws.

Traditional copper wire can only maintain a 400 Gbps signal over about 1.25 meters. Every time the per lane speed doubles, the reach is halved. Photonics are largely immune to this and can push that signal over kilometers, enabling infrastructure providers to scale their systems across entire aisles, while also reducing the power and cooling requirements of each individual rack.

Due to close in the first half of 2028, the all-stock transaction comes as AI infrastructure builders, including Nvidia and AMD, eye larger rack-scale systems, which will rely heavily on optical interconnects to stitch hundreds and even thousands of GPUs together into a single compute domain for training or inference.

At Computex 2026 in Taipei earlier this year, Jensen Huang predicted that Marvell's long-term investments in photonics would catapult the IP house to a $1 trillion valuation. That hasn't happened just yet - the company's market cap currently sits at just $180 billion - but it seems Marvell isn't above buying its way to the finish line by snapping up optics vendors before the AI boom sends their valuations skyrocketing.

As we've previously reported, the move to optical scale up networks has fueled a flurry of investment in optics, silicon photonics development, and manufacturing. Nvidia itself invested $2 billion into Lumentum just eight months ago to bolster its optics manufacturing capacity.

The limiting factor up to this point has been power. When Nvidia unveiled its first Blackwell-based rack systems in 2024, Huang noted the company specifically avoided optics as they would have added 20 kW of additional power draw to a system that already consumed 120 kW under load.

In the years since then, Marvell has invested heavily in next-gen optics tech. One of the most notable acquisitions came in February 2026, when the company paid about $3.5 billion for photonic interconnect developer Celestial AI.

Celestial's tech opened the door to more efficient optics integrations including near- and co-packaged optical interconnects that used a fraction of the power of conventional pluggable optics used in scale out networks.

Marvell's Lumentum buy will position the IP house as a one-stop shop for AI infrastructure providers' photonics needs. ®

The week after promising to cap new borrowing at $20 billion until the end of its 2030 financial year, Oracle launched plans to borrow all that in one go.

In notices to investors on Monday, Big Red said it is offering bonds worth $10 billion together with $10 billion in shares of its common stock.

It said it could use the money for working capital, capital expenditures, contractual obligations, acquisitions, dividends, stock repurchases, debt repayments and other matters. The most likely candidate is financing datacenters, as the tech giant is in the middle of capex binge which saw $55.7 billion spent in fiscal year 2026, which ended in May. The company plans to spend between $90 billion and $95 billion in fiscal 2027.

Last week, the company told investors it would limit new borrowing to $20 billion until the end of its financial year 2030, which ends in May that year. It promised to get more revenue from customers by charging for AI usage on top of seat-based subscriptions.

CFO Hilary Maxson also told an analysts' event that Big Red would look to borrow the $20 billion during the remainder of fiscal 2027, which ends in May, implying there would be no new borrowing for the following two years.

Maxson confirmed this was correct. "That's a real commitment. We don't expect other forms of structured financing or other things using our Oracle credit out in the market," she said.

In July, Oracle said it could face more than $100 million a year in financing costs to guarantee the power commitments behind a nearly 1 GW datacenter campus it is developing in Wisconsin with Vantage and OpenAI.

In October, Oracle reportedly sent a force majeure notice to a Blue Owl Capital subsidiary developing Project Jupiter in New Mexico. Bloomberg reported that Oracle was seeking to defer payments if the campus failed to come online as planned, rather than withdraw from the project. Oracle says the development remains on schedule.

In September last year, Oracle's valuation rocketed after it boasted $455 billion in RPOs, $300 billion of which turned out to be for OpenAI, all largely based around leasing datacenters for AI compute. ®

Ericsson is bringing drone detection to mobile networks, analyzing radio signal reflections to identify and track low-altitude objects.

The telecoms infrastructure maker says its Integrated Sensing and Communication (ISAC) tech adds detection to existing mobile networks, closing a security gap in tracking low-flying, fast-moving objects such as drones.

Following tests over the past few months, Ericsson says it will introduce the tech commercially in phases. Small-scale, sensing-only pilots with selected customers will start in 2027, followed by broader deployments from 2028.

CCS Insight research director Ian Fogg, said told The Register:

"While sensing can be used for other things — like tracking boats in a port, traffic patterns or weather conditions — drone detection [in civilian areas] is the most pressing need which cannot easily be solved using other technologies.

"Mobile networks are useful as drone detection solutions because their antennas are already covering low elevations, close to ground level where drones often fly, unlike traditional radar."

Although it is possible to use 5G networks for sensing, a standards-based 6G method could improve precision by letting different suppliers' base stations "interoperate", alongside a mix of client devices, tracking the "smallest and fastest moving drones in difficult environments," Fogg added.

"The new spectrum that should arrive with 6G will also make it easier to deploy sensing on wide channels which further improves detection accuracy," he said.

Small, low-cost drones have emerged as a persistent threat, or at the very least a nuisance, closing down airports due to safety concerns - the latest incident was at Jersey Airport over the Halloween weekend. Drone activity has also been on the increase near to military bases and other sensitive sites.

Dedicated sensor systems were deployed to cover sensitive sites, and in the UK, military personnel have now been granted the authority to neutralize any unwanted flying intruders.

Scaling detection across larger areas, however, is more challenging, Ericsson said.

ISAC transforms mobile network infrastructure into a distributed sensor array. Using sensing-enabled 5G radios, edge computing and signal processing, a cell deployment can detect objects and track them across the coverage area, complementing radar and other sensors rather than replacing them.

Ericsson says the tech works by having multiple perspectives from a number of cell tower radio sites. When one transmits a signal, it reflects off the drone and is picked up by multiple receiving sites. The network then analyses the reflected signals to locate the drone and track it.

ISAC requires sensing-capable radio hardware, dedicated compute resources, and software, says Ericsson. The radios have to meet specific Massive MIMO (multiple-in, multiple-out) antenna and spectrum requirements.

AT&T tested ISAC at trials in Arlington, Texas, and with Erillisverkot, a state-owned special purpose company responsible for national critical communications in Finland. In both cases, the tech demonstrated it was able to detect multiple drones and track their movements, according to Ericsson.

Data from ISAC - we're told - can be securely shared through standardized interfaces with authorized external systems, including security and command and control systems. ®

If you're a customer of cloud-hosted AI workflow platform Trigger.dev, take heed, because your environment variables may have been exposed in a security incident disclosed on Sunday.

Trigger offered little explanation in its Sunday security incident report, noting that its investigation had identified "exposure across a subset of organizations" and that an attacker had gained unauthorized access to environment variables belonging to those organizations.

Whatever exploit was used to gain access has been blocked, the company added, and affected customers have been notified of steps they need to take to prevent further security problems.

We likely don't have to tell The Reg readers how important environment variables are in SaaS and cloud services like Trigger. If compromised environment variables contain sensitive credentials, an attacker could basically end up with the keys to the kingdom, potentially gaining access to connected services and digital infrastructure.

For those unfamiliar with Trigger.dev, it's a platform for building AI agents and workflows in TypeScript, but here's the kicker: Trigger advertises itself as a durable execution platform that handles pesky things like retries, scaling, and keeping long-running tasks going through crashes and redeployments, while providing tools to monitor what those tasks are doing.

Trigger, the company's site notes, can be used to "build durable AI agents that survive refreshes, redeploys, and crashes," and is ideal for "long-running tasks with retries, queues, AI observability and elastic scaling."

In short, companies running their AI on Trigger probably want it to keep running, which makes stolen environment variables all the more dangerous.

Elastic scaling, automated retries, and, as Trigger notes, the ability to run lots of concurrent operations without timeouts could compound the consequences of compromised API keys, depending on what those credentials allow an attacker to access.

An attacker with stolen credentials could potentially interfere with a customer's connected services or workflows. Additionally, Trigger recently added webhooks to its offering, supporting services like Stripe, Supabase, and Slack, potentially expanding the range of integrations that could be affected if their credentials were exposed.

This could be pretty bad for anyone affected. Trigger has some fairly high-profile customers, with companies like fashion brand Emporio Armani, Pentagon-supporting AI acquisitions platform GovSignals, SaaS calendar platform Cal.com, and others all listed on the company's site. Trigger has not publicly identified which organizations were affected by the intrusion.

We reached out to Trigger, which declined to provide any additional information on the incident. A company spokesperson instead told us to wait for a full incident report to learn more.

Luckily for Trigger customers and those concerned about SaaS security now that AI is regularly coming up with new vulnerabilities in commonly used platforms, Trigger is open source and can be self-hosted, which might be a better alternative in this day and age than relying on a hosted service. ®

Victim organizations are losing millions of dollars in token charges as criminals hijack their corporate LLM access to steal expensive computing resources and AI services.

These fraudulent charges from resource consumption can exceed $100,000 a day, according to a Monday report by Google's Mandiant incident responders, who saw cumulative charges at some organizations hit $1 million before the victim companies revoked intruders' access.

"Whenever possible, threat actors want to use this technology to gain a scale advantage over defenders. They achieve that with agentic tools and by stealing tokens from unwitting third parties," John Hultquist, chief analyst at Google Threat Intelligence Group, told The Register. "Locking down these resources will not only protect organizations from runaway costs, but also protect the broader community from a deluge of cheap, autonomous attacks."

This tactic of hijacking access to LLMs is called LLMJacking, and it's been increasing steadily over the past two years. In one extreme case, documented in an August report from CrowdStrike, a token thief sent about 200,000 API requests in just two minutes.

In one case detailed in the report, a group Google tracks as UNC6993 illicitly gained access to LLM API keys, then generated additional keys and enabled services within the victim's cloud projects, racking up $100,000 in charges.

The threat intel biz also documented a different token thief subscribing victim accounts to Pay-As-You-Go agreements to access advanced models on the victim company's dime. The crook selected premium foundation models, launched high-volume inference streams, and used a "burn and rotate strategy" to bypass rate limits by creating and deleting short-lived service accounts while auditing the billing interface to monitor financial impact.

In yet another incident, an intruder executed more than 500,000 LLM API calls in less than eight hours using stolen third-party API keys to power a token routing service called Jet Night, according to a Monday report from Google.

"Given Jet Night's per-token costs to users are so far below typical market rates GTIG believes the service relies on stolen keys to operate profitably," the report says. "For example, Jet Night offers an unlimited token plan for approximately $60 USD per month with alleged access to frontier models, a consumer service level that would typically run between $100 and $250 per month."

Jet Night has also admitted to multiple outages on its public Telegram channel, which the Google threat hunters note is "likely due to stolen keys being revoked as victims discover their respective compromises."

As more criminals join the LLMJacking ecosystem, Google says it has documented a division of labor between the criminals that establish initial access to compromised cloud and AI services, and those that manage downstream monetization infrastructure, called "LLM Mixers."

In one case that Mandiant responded to in mid-2026, UNC6993 compromised a cloud instance through a misconfigured Service Account key and used this access to grant further access to a model API.

The crooks also spun up a session hijacking page purporting to be an application portal to join a group of LLM users. "This page tricks users into navigating to the legitimate LLM service website, and pasting their raw JSON payload containing an active access token, which was then posted to the backend API to bypass password and multifactor authentication (MFA) controls," the report says.

Google predicts miscreants will "increasingly prioritize LLMJacking" compared to other types of resource theft like cryptojacking. And to prevent cloud LLM service abuse, the security firm says defenders should prioritize identity and access management plus supply chain security, and use a centralized secrets management platform to rotate keys and secrets.

Also apply least-access privileges to users and service accounts, ensuring that they only have permissions necessary to complete their roles and tasks. And automate secrets scanning within codebases to identify and then remove any API keys and secrets added to source code. ®

One in ten employees at SUSE are facing the chop due to "automation and AI," according to a memo dispatched by management at the privately owned open source software biz.

In an email to the 1,900 strong workforce on Monday - seen by The Register - CEO Dirk-Peter van Leeuwen, the former Red Hat exec who joined SUSE in March 2023, confirmed the "difficult news."

"We are announcing a reduction of SUSE's global workforce of approximately 10 percent," he said. "These decisions are painful for our people, our culture, and our open source community, but they are necessary for us to adapt and build a sustainable, resilient SUSE for the long term."

Van Leeuwen says the business, bought by Swedish private equity provider EQT, decided to cut jobs after scrutinizing its "future strategy, required capabilities, and team structures." The rest of the organization will get a look at the fiscal year 2027 strategy on November 16.

He said in the message to staff:

"What is clear is that automation and AI are fundamentally shifting how software companies operate. AI capabilities are advancing exponentially, forcing us to radically reimagine our work. We have seen this trend already both inside our sector and in the wider economy. This means that the organizational design and cost base that SUSE is built on today will not carry us forward successfully.

"We are adapting. Across SUSE, modern tooling is already changing our daily workflows, enabling teams to move faster with less operational overhead and flatter reporting lines. The requirement for some existing roles has greatly diminished. Other roles will evolve over the next few years. This means we are accelerating investments in high priority areas, and deprioritizing others. These are hard, painful decisions to make, but necessary for our future."

As The Reg reported in 2025, SUSE doubled down on AI.

The specific roles at risk of redundancy were not outlined but SUSE claims it has considered where the people in those positions are "able to evolve, what they can stop doing, what they need to do more of - and decisions have been made on that basis."

Staff caught up in the redundancies will be informed by November 4, employee representatives nominated the following day, and one to one meetings will begin at the start of next week.

Van Leeuwen says in the internal memo that the job cuts are a "reflection of shifting organizational needs, not your talent or commitment. You have tackled hard infrastructure and open source challenges that power critical enterprise systems around the world.

"Any organization will be fortunate to have your expertise."

Back in September, we reported that the company was asking old hands if they would take "voluntary separation." It seems not enough people chose to walk through the door, so instead, they're getting thrown through it instead. SUSE "geckos" in the Czech Republic may be glad they joined a trade union.

Back in 2018, EQT bought SUSE from Microfocus for $2.5 billion. EQT floated SUSE on the Frankfurt Stock Exchange in 2021 with slightly disappointing results. A couple of years later, new CEO Dirk-Peter van Leeuwen took over, and later in 2023, EQT chose to take the company private again.

The commercial context is that in March SUSE's corporate owner EQT was reportedly considering selling it off again. ®

Cyberattacks on both Russia's and Ukraine's emergency alert apps resulted in fake messages about missile and nuclear strikes broadcast to the public.

Officials from both countries confirmed the attacks, which occurred at similar times, on Sunday evening.

According to public statements and messages sent via social media, Ukraine's alert app was hit first, with the country's Air Force quick to dispel nuclear fears.

"The Air Alert app has likely been compromised," it said via Telegram at around 2100 UTC. "There is no radiation threat! Stay calm, follow our messages on our channel!"

At around 0030 UTC on Monday, Russia's Ministry of Emergency Situations confirmed the country's app had also been abused, with fake messages sent to citizens' devices.

"Attention! The Russian Ministry of Emergency Situations' mobile app has been subjected to a massive cyberattack, allowing attackers to send fake information to the public," it said, also via Telegram. "Work is currently underway to patch the vulnerability and restore normal operation."

The ministry issued an update around 90 minutes later, saying the app was once again functioning normally, and pointing out that an attack on the app was not the same as a compromise of the state's national alert system.

"Citizens are alerted through primary channels: regional and municipal alert systems (sirens, loudspeakers), communication networks (SMS messaging), and television and radio broadcasts. The mobile app is merely an additional, supporting information service," it stated.

Ukraine's Ministry of Digital Affairs also issued a public announcement reminding citizens about how to receive emergency warnings while its "Alarm!" app was down. These include enabling iOS and Android emergency alerts, state-controlled sirens, and other official sources, such as the official social media accounts of Ukraine's government.

According to the messages sent to devices after the compromise of Ukraine's alert app, the fake warnings pertained to various threats, such as street fights in Kyiv, chemical attacks, and radiation hazards.

Alerts about the threat of nuclear fallout from official government warning systems would normally lead to panic, however, one of the earliest messages sent to citizens was written in Russian and referenced the "Oreshnik" missile, which was an immediate signal that the message may not be genuine.

Equivalent alerts sent to Russian citizens also warned of impending nuclear strikes "at all regions of Russia," referencing "P1-Sun," the name of a Ukrainian attack drone.

Oleksander Merezhko, who chairs the Ukrainian parliament's foreign affairs committee and noted the suspicious Russian-language alert, told the BBC the attack was almost certainly part of Russia's cyber war.

"It's a very unpleasant fact but it is part of cyber war and people in Ukraine understand that," he said. "It is possible and we are quite prepared for such possibilities."

Russia has not explicitly blamed Ukraine for the attack on its app, and official sources say the app is running as normal.

Ukraine's announcement of the alert app cyberattack came almost simultaneously with the official conclusion of a six-day ceasefire in Kherson.

Russia agreed to the rare temporary suspension of conflict to allow hundreds of Ukrainians to evacuate the Russian-occupied city of Olehsky.

Russia did not agree to a proposed extension of the ceasefire, which officially ended at around 2100 UTC - around the same time Ukrainians began receiving fake warning messages. ®

Microsoft has discovered once again that few things strike fear into the heart of a software engineer quite like a government deciding to fiddle with the clocks.

Parts of Canada stopped observing seasonal clock changes on November 1, leaving affected areas of British Columbia, Alberta, and Manitoba permanently on UTC-07:00, UTC-06:00 and UTC-05:00 respectively. Unfortunately, Microsoft's sprawling software empire hasn't entirely caught up.

The company has confirmed problems across a chunk of its software estate, leaving some customers facing incorrect times, broken schedules and a collection of workarounds while Redmond catches up.

Outlook users have been left with the most homework. Switching to the new Canadian time zones won't automatically update existing calendar entries, meaning that meetings scheduled after November 1 are potentially an hour out of whack.

Microsoft advises users to change their Outlook calendar time zone, then update upcoming events and meetings to use it. For meetings involving other people, only the organizer can make the change.

Recurring meetings come with an additional treat: changing the time zone resets any previously moved occurrences, which then have to be recreated. Anyone with a calendar full of recurring meetings may wish to put the kettle on first.

Microsoft is working on an Exchange Online PowerShell cmdlet to remediate non-recurring meetings, but according to its guidance, that's not due until later this week. A fix for recurring meetings remains "in development," with an ETA yet to be confirmed.

Things get stranger with Power Automate and Fabric, where Microsoft's temporary fix involves pretending you're somewhere else entirely. Depending on the product and province, that could mean Saskatchewan, Arizona or, for Manitobans, Bogota. Because apparently time zones weren't confusing enough already.

SharePoint On-Prem customers, meanwhile, have to wait until November 10 for the correct time zones to arrive, while Power Pages users could see server-formatted dates appear an hour out until a service fix lands. Microsoft's current milestone for that is December 12.

Admins dealing with the fallout aren't exactly delighted. One sysadmin on Reddit said Microsoft's guidance had been "terrible" and claimed to have written roughly 900 lines of PowerShell to change users' Outlook settings and manually adjust calendar events. Others described switching systems to Saskatchewan time while waiting for Microsoft's new zones to settle in.

Microsoft published its guidance on October 26, less than a week before the change took effect, and was still adding affected products and workarounds four days later. Meanwhile, admins were urged to test calendars, recurring meetings, and schedules before the November 1 switch, leaving little time to find out what might break.

If any of this sounds familiar, it's because time has caused trouble for Microsoft before. In 2022, Chile delayed its switch to daylight saving time by a week. Microsoft responded by warning that Windows clocks could be wrong, scheduled tasks could run at unexpected times, and timestamps on files, transactions, and logs could end up an hour out.

Four years later, governments are still messing with time and software is still apparently objecting. ®

The cyberattacks on commercial refrigerators and freezers in Europe continue, as a slew of grocery stores in Belgium, Ireland, Luxembourg, and Sweden report widespread food wastage as a result of miscreants raising the temperatures to highs of 50ºC (122°F).

In Belgium, around 40 Delhaize stores were affected after the attackers reportedly switched fridges and freezers into defrost mode overnight, between Sunday and Monday, resulting in stock issues.

Diego Louwers, a Proxy Delhaize store manager in Overijse, near Brussels, told RTL Info that the consequences of the cyberattacks were "catastrophic," forcing the store to throw out the entirety of its fresh and frozen products.

"We can't have anything worse in our line of work," he said. "We have to throw away all of our merchandise, and it's both heartbreaking and very complicated operationally."

A Delhaize spokesperson said it took steps to ensure stores remain stocked, although fresh and frozen products are "temporarily somewhat limited."

"On Monday, 2 November, we identified an IT incident affecting the refrigeration systems in several dozen of our stores," they said.

"We immediately took the necessary measures to keep our stores operational and ensure that our customers can continue to do their shopping as usual.

"In the affected stores, the range of fresh and frozen products is temporarily somewhat limited. Our distribution centres are operating normally, and there has been no disruption to deliveries to our stores.

"We are, of course, continuing to ensure that our products are stored and presented under optimal conditions, so that we can provide our customers with the quality and freshness they have come to expect from us."

Around 15 Carrefour stores are also reportedly affected, as are sites operated by Intermarché, Spar, and Shop & Go stores located at Q8 service stations.

Belgium's De Laet & Van Haver chain of butcher shops reported temperature rises in its appliances too, but followed up with a statement on Monday saying there was no impact on its produce, and all stores remain open.

Simultaneously, similar issues have affected Swedish grocery chain ICA, which confirmed attacks on a "small number of stores."

"We can confirm that we have been informed of an incident overnight affecting a small number of ICA stores operating outside ICA's central network," a company spokesperson said.

"The incident impacted the stores' refrigeration and freezer systems. ICA is currently supporting the affected stores in managing the situation."

ICA did not respond to questions about the scale of food wastage nor when normal operations are expected to continue.

One store, ICA Nära in Böda on Öland, said via social media that it is open today but has no fresh or frozen produce to sell.

"Unfortunately, we are one of the stores whose refrigerators and freezers have been hacked," it said. "We are open as usual, but unfortunately we have no refrigerators and freezers to sell.

"We have no status on when the refrigerators and freezers are expected to be up and running again."

The store manager, Mats Olsson, told SVT that he arrived to work on Monday morning and was greeted by a puddle of water and freezers set to 55 degrees C - a tale echoed by other managers across the country.

Coop stores are affected by similar attacks too, with six locations in central Sweden experiencing fridge-freezer issues, according to a company spokesperson.

Likewise, at least 13 SuperValue stores in Ireland are also affected, with fresh and frozen food being thrown out, according to parent company Musgrave speaking to RTE.

The news comes less than a month after various grocery chains in France resorted to discarding bucketfuls of produce after their appliances were remotely defrosted by cybercrooks.

Carrefour, Coopérative U, Intermarché, Monoprix, and others were hit with cyberattacks in mid-October, and hundreds of stores across the chains were reportedly impacted.

At least some of the affected stores were running outdated firmware on their Danfoss refrigeration controllers, and shortly after the attacks were confirmed, CERT-FR published an alert pertaining to numerous Danfoss AK-SM and AK-SC controller compromises.

Other retailers, such as Coopérative U, confirmed they were not Danfoss customers.

It remains unclear if the latest attacks across Europe used the same methods as those used for the disruption in France last month.

None of the companies that spoke to The Register on Monday responded to questions about who was behind the intrusions. ®

It's official: Anthropic is going public, and boy, are our eyebrows tired from poring over hundreds and hundreds of pages of questionable claims.

You can listen to the latest episode of The Kettle right here on this page, as well as on Spotify, Apple Music, or YouTube. Those platforms also let you subscribe to The Kettle, so you are always notified when the latest episode goes live.

Anthropic is going to need to do a lot to make its massive IPO more than just a flash in the pan, as systems editor Tobias Mann and senior reporter Tom Claburn tell host Brandon Vigliarolo on this week's episode.

For one, the company has nearly a trillion dollars in compute commitments to pay for, which is going to mean it'll be burning through more cash than a datacenter gas turbine turned up to 11. In addition to its massive spending for the foreseeable future, there's also the fact that much of its computing power relies on customers who are also competitors.

In short, we're skeptical, but you probably already knew that. It's El Reg, after all, and we're not just taking what the AI industry claims at face value - if it looks like a house of cards, it probably is.

A lightly edited transcript is below:

Brandon (00:03)

Hey everyone, and welcome to the latest episode of The Register's Kettle podcast. I'm your host, Reg reporter Brandon Vigliarolo, and this week we're focusing on what I think our contributor Carly Page aptly described as the peak of the AI hype cycle we've been in for the past few years, and that's the public release of Anthropic's initial public offering filing. With me to talk this over is our systems editor Tobias Mann and senior reporter Tom Claburn, both of whom I'm sure have plenty to say about this matter.

Welcome guys.

Tobias Mann (00:34)

Good to be here.

Tom Claburn (00:34)

Good to be here.

Brandon (00:36)

So, I'm sure we've got a lot to cover. 614 pages is a lot to go through. Tobias, I understand you were combing through this thing with a fine-toothed comb. Talk us through the hard numbers before we get into some of the more wobbly parts of this IPO.

Tobias Mann (00:56)

Sure. At face value, going into any S-1 filing, it is a kind of mess of risk mitigation, kind of covering your butt. They all pretty much read like a doomsday prophecy.

Brandon (01:18)

Right.

Tobias Mann (01:18)

The S-1 definitely did that. But on the numbers side of things, the takeaway for the market was kind of interesting because there were some people looking at the numbers and going, "OK, this actually seems reasonable." Their operating losses were only about 700 million dollars.

Brandon (01:40)

I think on 16.3 billion net [revenues], right? That was what they reported.

Tobias Mann (01:46)

Right.

Brandon (01:47)

Now they also mentioned, I think, and I was going to ask you about this, $80.7 billion total losses, but I understand that was an accounting adjustment of some kind. Did they mention in the IPO what that meant specifically?

Tobias Mann (02:01)

Not specifically. It's a little fuzzy, and it's always kind of like this.

Brandon (02:07)

Of course it is, right?

Tobias Mann (02:09)

Anthropic's business is extremely cash heavy, and so they had to go to partners and say, "Hey, we need capital in order to pay for compute to train these models. We can either take it in debt or you could also get repaid in valuation, our equity." And so I believe a large part of that loss is they're trying to write off a lot of the equity that they're going to have to pay off when they do go public. And so I do believe that's the lion's share of that loss. Because it does look huge, but when you think about how companies go public and how they raise capital as startups, it's not really that surprising. It's just a huge number because Anthropic is arguably the largest IPO in US history at this point.

Brandon (03:02)

Right. But bottom line is their actual day-to-day business losses were $693 million....It was $1.2 billion, I think, for the first half of the year or same time last year or something like that. So it's gone down.

Tobias Mann (03:17)

Yes.

Tom Claburn (03:17)

And they're trying to go out at two trillion now? Is that the number?

Tobias Mann (03:25)

Well, the number kind of depends on whether they're going to be able to sustain this. I was going to mention for the first time ever, their revenues outpaced their compute expenses two to one. This has never been true in the company's history, and it's one of the reasons that people are feeling more confident about this. The issue that really comes from all of this, and what has gotten missed somewhat in the more rosy interpretations of the S-1, is that they also are signed up to have to repay upwards of 800 billion dollars and counting of compute commitments out through the next two decades or so.

Brandon (04:12)

You outlined in your story just a whole host of people they owed billions of dollars to for compute. Like you said, it would be great. That $16.3 billion in [revenue] would be great if those compute costs were going to hold steady, right? But they're not. They're going to keep climbing and there are a lot of bills that are going to be due. So, profitability is relying on making more than that nearly trillion dollars in compute obligations, and that's a big if.

Tom Claburn (04:43)

I mean, the big question mark is in terms of what the technology does for inference costs, because I think a lot of this is betting on being able to get the same hardware that is much more efficient down the road and maybe not drop your prices proportionately and profit on the difference. But it remains to be seen whether there will be enough competition to just have the price go do with the increasing cost. And you see that a lot. I think the inference costs have dropped pretty significantly over the year, on average. I mean, it's really hard to calculate inference costs because there are so many different factors and it's really situational depending on the workload and how much caching is going on and things like that. But overall it's a lot cheaper to do inference now than it was a few months ago.

Tobias Mann (05:34)

Per token, yeah.

Brandon (05:36)

But again, I feel like if you're betting this $2 trillion IPO on the fact that we're going to keep getting better and cheaper at inference, right? I mean, I know Moore's Law is kind of a thing of the past, but at this point, what if we run up against another compute bottleneck and inference stops getting cheaper before they need it to?

Obviously any IPO prospectus is one big document about risk, right? Like, "Hey, you might make a ton of money investing in this company, or we might go complete belly up and you lose everything," right? That's always a possibility.

Tom Claburn (06:15)

Right. And they still haven't demonstrated that the use case for AI is wide enough at volume that they claim it is. I mean, in certain sectors it's pretty obvious now there's going to be AI for a long time. Coding, everyone kind of agrees that works and we know how to manage that. And then there will be adjacent fields to that. But once you start getting into more general office work and regular business tasks that are not super-compute heavy, I don't know. It's a good question. We'll see if people are going to use it that broadly, but I just see it being overly optimistic.

Tobias Mann (06:59)

The big question I have right now is: Is there enough user demand to support even a $200 billion a year revenue stream? Because even going off just commitments that they can't get out of made in the first half of this year, that's $417 billion running out through about 2033 to 2036.

Brandon (07:32)

I think the upper end of your estimate was a total, including newer commitments, of about $807 billion. At the lower end, $782 billion at the lower end. That's not that much of a difference when you're looking at hundreds of billions of dollars.

Tobias Mann (07:46)

But if they can't make a profit with a two to one revenue-to-compute ratio — if they have to make twice as much in revenue in order to pay for their compute and still not break even — you extrapolate this out and they're going to have to be making $200 billion plus, assuming no additional compute contracts are made during that period over the next year, which is definitely not going to happen. They've already said they're going to keep making compute contracts. So this number just keeps getting more absurd the further out you go, unless inference magically gets dirt cheap, which we haven't seen.

Brandon (08:33)

Right. And it could cap. You know, I think also you made a really good point in your story, Tobias, when you mentioned that some of their biggest customers are also some of their biggest competitors. They're leasing space from hyperscalers, they're making deals with people to host their compute. But these are also companies that are operating their own AI models. And you might end up in a situation where they're like an MVNO, which, if anyone's not familiar with a mobile virtual network operator, is cell phone companies that are not T-Mobile, Verizon, or AT&T. Anyone who's not those big three in the United States is operating a virtual network that piggybacks on someone else's infrastructure. And when that infrastructure is busy, you get less priority. So we could see them run into a similar situation there, I imagine.

Tom Claburn (09:26)

There's an interesting trend that another piece I'm working on later today touches on. Some of the AI gateways like OpenRouter are now starting to move away from just focusing on price metering and looking towards things like fastest time to tokens and all the quality of service metrics in terms of the way that they price and deliver the models, because inference pricing has become very commoditized in terms of what they're dealing with. They're not necessarily serving all the frontier models at very high prices. They're dealing with a lot of open weight models. And if you just focus on the lower end of the market, there's not a lot of margin there. So it makes sense to go, "Well, maybe if we deliver this faster or have better reliability or service, we can move up the price and get a little profit." And I think everyone in the industry is starting to realize that the music is stopping and they have to take a seat at some point and make the business model work. So there's going to be a lot of pressure on that. And that's also showing up in things like the licensing of models. A lot of those open weight models have seen their licenses get more restrictive over time as companies realize, "Well, maybe we don't want someone competing with us with this thing that we've made." So I think there's a lot..

Brandon (10:48)

That reminds me of Mistral just releasing the open weight version of Large 4, "Le Chonk". And the licensing model basically said if you intend to use this for any down-market purposes, you've got to sign a fairly restrictive agreement with us so that you're not just using this open weight model without us benefiting from it, right?

Tobias Mann (11:15)

A lot of the big Chinese models, like Kimi K3, I believe were among the first to move back in this direction, basically saying, "Hey, if you're just going to deploy this internally, we don't care." If this is just for your enterprise, fine, but if you're going to be selling an API service like OpenRouter around it, you gotta cut us in. So there is definitely that. And Anthropic in their S-1 - again, every S-1 reads like a doomsday prophecy. And this S-1 of the things that they raise: open models are a threat on multiple levels, one of them being that a lot of their customers are startups sensitive to economic downturn. So those customers will likely be the first to jump ship to cheaper alternatives that are maybe not as good, but they can afford if the economy runs into any issues.

Brandon (12:25)

I wonder if that'll happen or not, right?

Tom Claburn (12:28)

Was it Elon Musk who was tweeting about the end of the neoclouds recently? With financing getting tighter, a lot of the neoclouds living on borrowed money are going to find it harder to sustain their loans and are not going to survive this. So there will be a shakeout at some point. Does it hit the big companies? I don't know. All the cloud providers look pretty happy because they get business regardless. Even if Google's AI models are not first in people's minds, they're still serving all of this stuff.

Brandon (13:08)

We've said that on multiple episodes, and in multiple articles we've written, right? Basically, if the AI industry does implode, Google's going to be fine, Amazon's going to be fine, Microsoft's going to be fine; these guys have other businesses. Anthropic and OpenAI, not so much. That's all they do. And in many cases they're renting compute space from these other providers. So they might take a bit of a hit there, but by and large, they're going to be okay. They've got compute, they'll turn it around, they'll find other people to use it. Which is interesting, from your article, Tobias, you kind of summarize that the entire pitch this IPO makes is that bigger is better, right? But as we were just talking about, there are a lot of reasons to say that's probably not the case. It makes me wonder if they're still living in the token-maxxing era, whereas I just wrote a story this week about AWS telling its engineers to back off, stop using so many tokens, be more judicious, use AI well, don't just use it. And Microsoft said this months ago in August - they were saying tokenmaxxing doesn't factor into how they want to do business. So you've got two very big consumers of AI basically saying, "Y'all need to slow it down and be more purposeful." So, bigger isn't necessarily...

Tobias Mann (14:33)

Anthropic's business is predicated on the idea that more compute means more revenue. However, the disconnect is: Can revenue scale fast enough? Is there enough user base? Is there enough token demand to support the scale they need to cash the checks they've already written? And it really looks like a house of cards. They warn about this in the S-1 - they have to - that if their revenue can't keep up with their compute commitments, they could default. And that default could turn into a cascade where if they can't pay one bill, they can't get enough compute to generate revenue necessary to pay the next bill, and it all falls apart.

Brandon (15:28)

If it happens to Anthropic, the other AI frontier labs are not far behind, because that's going to be a symptom of a larger problem. And that's that people are basically saying what a lot of non-AI companies or companies that do AI along with other things, like AWS, are saying: We don't need these massive frontier models for the vast majority of business purposes. They're not necessary.

Tom Claburn (15:54)

Right. A lot of the enterprise AI products you see now are about trying to figure out ways to reduce that cost, too.You have some sort of gateway or harness system that looks at your jobs and routes them to an appropriately cheap model so that most of the time you're not using top-tier intelligence; maybe you're using a decision model, a regular classifier, or something like Gemma to do language translation. You don't always have to use that expensive model, and that price pressure is going to hit them, too.

Tobias Mann (16:37)

To Tom's point, Cadence Systems, which does a lot of electronic design automation and circuit board layout, just rolled out a new product line called Maestro that's built around Nvidia's Nemotron open-weights models. The idea is that they had been using proprietary APIs to power their agents that assist engineers with PCB layout and chip design, but those agents generate a ton of data that could be used to train and fine-tune open weight models. So their latest pitch is, "Hey, we know our customers like this, but it's getting expensive to continue using these proprietary models from OpenAI, Anthropic, or Google. So we're going to give you a way to take an inexpensive model, use the proprietary model to generate training data that we can use to fine-tune this open model, and then you can decide how much of the work you want done on the expensive proprietary models versus the cheap open models. You can deploy those models on-prem, in our cloud, or on your preferred cloud. Just use our agent platform and pay the subscription fee."

Brandon (18:03)

I can't imagine most businesses are going to fork over all this money to proprietary model providers when they don't need those resources. It really makes me wonder how much of this IPO is forecasting the bursting of the bubble. I feel sometimes like we talk about the AI bubble bursting like doomsday prophecies: "It's happening now!" But there have been more signs recently. Chinese AI stocks have been hammered recently. There have been recent earnings reports from American companies that have not looked Yeah.

Tom Claburn (18:57)

Google was hammered; they lost like $500 billion in stock value based on the fact that their search was underperforming

Brandon (19:07)

That's that one I wasn't able to think of, that minor little blip, right?

Tom Claburn (19:11)

AI overviews not monetizing as well. So there's a lot, yeah.

Tobias Mann (19:15)

A problem they created for themselves.

Brandon (19:18)

Google creating its own problems? What?

Tobias Mann (19:24)

I would agree that Anthropic's survival depends fairly heavily on pushing the frontier forward. In order to do that, it needs a disproportionate amount of compute that it may not be able to afford on a continuing basis, especially if Microsoft, Google, and all of these cloud CSPs that provide compute for them and are among their biggest customers decide, "We don't really need cloud for this. We've reached a point where our own open-weights models or small proprietary models are good enough for drafting emails or generating code bases for our customers."

So they're going to start buying fewer tokens from Anthropic or OpenAI when they eventually IPO. They're likely to go through very similar challenges because they're also getting most of their compute from companies like Oracle and Microsoft at this point.

Brandon (20:38)

Which was one of the questions I wanted to ask as we wrapped up: Do you think OpenAI is going to be in any better position once their IPO comes out? Because neither of them owns a lot of their compute, and OpenAI's attempts to build its own compute seem to have fizzled out a bit. Could they be in a better position or is it going to be more of the same?

Tobias Mann (21:05)

I don't know with OpenAI whether they're going to be able to overcome the same risks that Anthropic has. It does feel like Anthropic is sidestepping a lot of hard questions by pushing forward with their IPO now.

Tobias Mann (21:24)

With Anthropic going public this year, if OpenAI goes public next year, I don't think they can avoid the scrutiny, because the world is going to have woken up to the risk factors associated with these giant model houses. And I think they're going to face a much harder time with this, even before they get there.

Tom Claburn (21:52)

OpenAI has the distribution of ChatGPT, which is demonstrating itself to be a useful tool that people like to use. But that's not where their monetization gold mine is going to be, unless they become the Google of AI search, and Google is already trying to get to that space, too. So I don't know how they're going to make it into the enterprise business and get the margins they need to pay back all the money they've taken.

Tobias Mann (22:31)

OpenAI also has a different business model than Anthropic, and to a pretty hilarious degree, if you recall the advertisements during the Super Bowl about injecting advertisements into GPT....

The point being that OpenAI's revenue streams may be sufficiently different to insulate them from this. They have been trying to position themselves as an AI advertising company as another revenue stream, which Anthropic has flatly rejected. So it's possible that OpenAI's business model is sufficiently different by next year.

12 months is a long time in this industry. If Altman takes them public this time next year, the conversation might be very different. Having said that, Wall Street's eyes are going to be open to a lot of the risk factors in a way that Anthropic is sort of sidestepping right now by being the first really big model house to go public.

Brandon (23:54)

Which is honestly kind of surprising to see OpenAI press the brakes while Anthropic presses the gas. I'm assuming they saw an opportunity to get out the gate first and avoid answering the hard questions. Because OpenAI basically said, "Hey, we're going to pause our IPO while we assess these risks," given they've been in the doghouse regarding AI behaving badly this year, giving Anthropic that opportunity to scoot past and not have to deal with these things.

Tobias Mann (24:26)

I think the other tell with Anthropic is actually that two-to-one compute ratio, because that's the first time that's ever happened. And it will be very telling if in their first or second public quarter that just disappears. So the timing might be related to the fact that they can now claim their cash flows are on a positive trajectory towards being cash flow positive and moving towards profitability. OpenAI delaying their IPO, I suspect, is largely because they need to get to a similar point. Otherwise, people are going to look at this and go, "This has no possibility of ever being profitable." At least with Anthropic's S-1, halfway through 2026, it looks like they could be profitable. Whether that's true in in their first public quarter is another matter until

Brandon (25:23)

To close us out here, I think this statement from your article, Tobias: "The only way to justify this kind of spend and risk is if you believe the upside is enormous, like the biggest economic transformation in the history of humankind enormous." And that raises my final question for you both: Is it going to be worth it? And what's the chance that Anthropic is going to flame out spectacularly in the next year or two?

Tobias Mann (25:49)

Tom, I'll let you take the first crack at that one.

Brandon (25:51)

Ha ha ha.

Tom Claburn (25:51)

Yeah. I mean, I don't know how spectacular it will be. I think it's gotten to the point where it's sort of too big to fail, but I could certainly see some desperation moves where one of these big AI companies strikes a deal with Apple for distribution, because they really need the kind of money that Google was paying Apple for, or sorry, that Google is paying Apple for for distribution. And I can see something like that happening because they really need revenue. I don't really see it blowing up because it's everybody. So they're going to figure out a way to do the least harm if things go downhill. And that may mean parting out part of their businesses. Who knows?

Tobias Mann (26:35)

I would tend to agree with Tom. I think about what happened with the dot-com boom as an indicator. The situation is quite a bit different because the AI industry has wormed its way so deep that it's not the surface-level situation we saw with the dot-com boom. There was a ton of fiber laid during the dot-com boom that didn't get utilized until much more recently, but it did get utilized. So I think AI is probably bigger than the internet and going to go down in the history books as the next industrial revolution, in much the same way the internet felt like that.

After the dot-com boom, it was big, but it wasn't as big as people prescribed the value to. It took another 10 years for it to actually become the thing everybody wanted it to become. It eventually rose to that valuation, but it took much longer than the market could sustain. And I think we're probably in for that.

Brandon (27:57)

So in other words, don't buy during the IPO. Don't buy high and hope to sell higher, because you might have 10 years to wait before you get back to that IPO valuation or above. Who knows? Maybe it'll go faster. No matter what happens, there's a good chance we're going to be right here to talk about it because this sure can fill a lot of podcasts and articles. Digital ink is going to be spilled on this topic for years to come. We hope you'll join us to hear all about it on the Kettle. Thanks for listening. ®

The FBI has blamed North Korean hackers for the record-breaking $6.3 billion theft from cryptocurrency exchange Binance, pointing the finger at the notorious Lazarus Group as the exchange scrambles to replace stolen funds and reassure customers.

In an alert published Saturday, the agency attributed the October 29 attack to TraderTraitor, a North Korean hacking operation linked to Lazarus, which has made a lucrative career out of relieving crypto exchanges of their digital assets.

Pyongyang was already in the frame before the FBI weighed in. Blockchain intelligence firms TRM Labs and Elliptic had linked the theft to North Korea, with TRM reporting that most of the $6.36 billion haul disappeared in about 20 minutes. Elliptic, which puts the total closer to $6.43 billion, says it has verified that $2.85 billion of the stolen assets have been frozen or seized.

According to the FBI, the hackers have started swapping their loot for Ethereum, with investigators expecting them to move the funds through decentralized finance (DeFi) platforms and instant swap services to make the money harder to trace.

The agency has published a list of cryptocurrency wallet addresses linked to the theft and is urging exchanges, blockchain analytics firms, and other virtual asset providers to block transactions involving the stolen funds.

The FBI's attribution comes days after Binance admitted that hackers had made off with roughly $6.3 billion in cryptocurrency. According to the exchange, the trouble started with a compromised DevOps engineer's laptop, which gave the intruders access to a service container and allowed them to authorize withdrawals they had no business making.

Binance has since resumed customer withdrawals, which it suspended after the attack, and says it has now replenished about 95 percent of the stolen assets. The remaining tokens are proving harder to replace because of their limited liquidity.

The exchange has committed billions of dollars from its own reserves and emergency funds to cover the losses, with co-CEO Richard Teng promising that customers won't be left footing the bill.

That hasn't stopped users from heading for the exits. Binance recorded approximately $3.64 billion in net outflows after withdrawals resumed, although the company said this represented only around 2.1 percent of its reported assets.

Replacing the stolen cryptocurrency is one thing. Getting it back from North Korea is another, and investigators face the considerable task of tracking billions of dollars as the hackers attempt to move it beyond their reach.

TraderTraitor has form when it comes to raiding crypto exchanges. The North Korean crew has previously been linked to a string of cryptocurrency thefts, including the $308 million attack on Japan's DMM Bitcoin exchange in 2024. The group has previously used fake job offers and recruitment exercises to trick employees into handing over access to their employers' systems.

As we reported here on Friday, crypto exchange Bitget has now confirmed that the cyberattack where hackers drained $387.5 million worth of digital assets from the exchange's wallets also has all the hallmarks of a North Korean operation. Its CEO did not go into too much detail, but did say that "IP behavioral patterns and on-chain signatures" suggest it was Kim's cronies at work.

This isn't Binance's first expensive encounter with hackers, either. Back in 2022, attackers exploited its BNB blockchain bridge and made off with roughly $566 million in tokens.

Four years later, the sums have grown considerably. Binance may have the reserves to make its customers whole, but the attackers responsible for the latest raid still have billions of dollars to launder. ®

From the department of "pointless but still cool" comes word of some kit being repurposed as a terminal for AI agents.

AWS developer advocate Darko Mesaroš has connected a teletype to a modern command-line interface to revel in the delight of interacting with today's AI technology via a combination of a DECwriter and fanfold paper, or the display and keyboard of a computer whose heyday ended 40 years ago.

Mesaroš' project involved AWS Kiro, a Command Line Interface (CLI) and agentic coding service. We would, however, contend that this is more about seeing what retro hardware can be made to do.

Two problems needed solving: getting data to and from the teletype, and formatting the output to work on a piece of kit from another era of computing. He used Hayes modem emulation to deal with the serial connection, and Agent Client Protocol (ACP) for working with Kiro.

Mesaroš used a WiModem232 for the former. The device emulates a standard Hayes-compatible modem and, according to Mesaroš, "Folks initially used this to connect Commodore 64s to BBSes in the modern era" (he started his computing journey with a Commodore 64). The latter, getting the Kiro CLI to work on a Teletype, was a little more complicated. Kiro's CLI isn't really suitable for a teleprinter capable of 30 characters per second, after all.

A simple Rust application took Kiro's output (obtained via ACP) and turned it into something that a teletype, or dumb terminal, could handle.

As we said, seemingly pointless, but making it work required knowledge of Rust, competence with containers, an understanding of ACP, and more than a little creative thinking. So, while one might argue it is pointless from a technical perspective, it is wonderfully educational otherwise.

The repository is available and focuses on Kiro for obvious reasons (Mesaroš is an AWS developer advocate after all), but in a chat with The Register, Mesaroš said the concept was portable. He said, "Any coding assistant or any harness that supports ACP, this should work, right? I built it for Kiro, but it uses the standard ACP protocol, so I could easily plug in Claude Code or Codex. Anything that basically can serve as an ACP server should be perfectly fine.

"Blessed be the standards, right?"

There might be some issues. "The biggest problem with this thing is that it doesn't really handle packet drops at all," he explained. "Like, if you drop packets, they're just gone. There're no retries. It's just spitting information out."

And then there is buffering. Or the lack of it. Mesaroš said, "There's actually no buffering happening here. Literally, it just comes down through the fire hose through this serial device, and the bottleneck itself is the serial device."

Mesaroš had toyed with using the GPIO pins on a Raspberry Pi, but settled on the serial device instead, since everything worked "out of the box." "It's kind of cheating," he admitted, "but it's also kind of good."

Mesaroš collects vintage hardware, with a stash that ranges from home computers made by Commodore and Sinclair, to a 1974 Data General Eclipse minicomputer and a plethora of Sun hardware. And, of course, an old Decwriter. He is also a dab hand with the soldering iron, having learned the skills to keep old hardware operational. Unlike today's equipment. "I have this big peeve over modern computers, where you just can't fix them," he said, "and it feels like such a waste of resources that when my MacBook dies I just have to drill a hole through it so nobody steals my data, and throw it away."

As for Kiro-TTY - or accessing a coding assistant or agent over serial using a decades-old device - he wrote on his blog, "This absolutely serves no purpose."

However, it did provide a novel way of working out how the ACP protocol worked. Perhaps not so usefully, it also taught the developer how to debug serial connectivity.

"And in general it helps me stretch what I can do in the current age of AI coding, and keeps my curiosity alive."

The destination might be pointless from a modern perspective, but the journey getting there is anything but. ®

Capitalism can seem so perverse. Take Microsoft, which is celebrating record financial performance by inviting 60,000 employees to consider continuing their careers outside the company. Voluntary redundancy may seem to make sense when a company's running out of cash, but Microsoft is dripping with the stuff.

It was a very good fiscal '26 at Redmond. Its Azure business is up over 40 percent, it's has 30 million paid seats on CoPilot, revenues are through the roof, and it has so much moolah it returned $10.2 billion to shareholders in the form of dividends and share repurchases.

This latter move probably soothed CEO Satya Nadella, who suffered, due to the vagaries of C-suite reimbursement and incentive packages, a 6 percent cut in his pocket money. But as he's also a stockholder with about 0.01 percent of the company, his share of the cash return presumably eased the pain.

He won't have been offered redundo. Jenkins, the large wheelbarrow if you please.

Microsoft, then, can afford all its employees, each who contributes about $1.5 million to SatNad's bottom line. The "massive efficiencies and ROI of AI" make a great line for a company flogging magic LLM juice, but the evidence remains elusive. AI is, of course, still the reason. Nothing else matters these days. It's just that it's not today's AI.

Among all the fulsome figures, analysts noted that Microsoft's capital expenditure wasn't as spendy as expected. Quite a lot has been pushed into next calendar year, not because it's not there to be spent - Microsoft could build a 1:1 scale model of the Vatican filled with robot priests if it wanted - but because so much it wanted to mash the Buy Now button for was "Out of stock. More expected soon."

This is bad. MS added 88 datacenters in its most recent full financial year. That's not enough datacenters for a cloud business expected to grow by half again in 2027, and hugely not enough to do the other thing. Lags in the cadence of deforestation, dewatering and detonation of the planet can't be tolerated. Thus there are contracts signed to get things back on track PDQ, with mucho, mucho, mucho sponduiks.

Microsoft is hinting, those selfsame analysts say, that it will have to go to the debt market and go into "negative free cash flow" territory in 2028. In other words, as well as spending its own bank, the company will be taking out massive loans.

This is why Billy and Bhavya Dev have to go. The cost of borrowing depends on the financial health of the borrower, and that boils down to how much profit you're making from the money you earn. A big round of redundancies, which are very costly in themselves, sits on the books of the year they happen. The hole in expenditure left by the vanished salaries lives on. The capital must flow.

Will this work? Last week, we listed the reasons why it might not — price competition between rivals and increasingly powerful local and open weight models, supply side shortages, you know the litany by now. This week is Groundhog week, with fresh stories singing the same songs from different choirs.

Any sensible soul, especially developers sitting on the sharp end of AI's forcible insertion into their working lives, will know how uncertain the job market is. Voluntary redundancy is a very high stakes bet, and the game is very different if you're in an organization that's going downhill and may need to get rid of you later with far less cash available than if you're nominally making $1.5 million in a place whose biggest problem is that it physically can't move into a bigger shell.

On average, only 5 percent or so of those offered voluntary redundancy take it, which Microsoft HR will have factored in. It's cold outside and snug within, so that may be on the high side. The biggest knock-on of such schemes is cultural, with only the very confident moving on out, a disproportionate loss of talent and role models. Hey, the cost of doing business, right?

For those who do decide to leave and watch the fireworks from outside the tent, part of the deal is career development resources. Sounds great, if anyone knows what technical careers will look like if either AI works and does indeed do all the work, or it fails and the global financial system throws a humongous wobble. Or anything in between.

So, here's an exercise in very dark humor quite fitting for the Halloween season. Ask your favorite AI what careers will be available for coders made redundant by AI. It will give you a long and plausible list. Then ask it - and yourself - why AI won't be able to do any of those jobs too. Buying a truck and a wrench and calling yourself a plumber is probably safest. Nobody will ever want AI to fix their toilet. ®

As robotaxis continue trials in London, an independent research group thinks the UK needs to answer some serious questions before heading any further down the road.

To kick off, the sovereignty question, or "Will Britain become dependent on autonomous vehicle services for any key activities and what fallback options will exist if a supplier fails or withdraws?" as the Institute for Driverless Transport (IfDT) puts it.

Another question concerns employment: "To what extent does professional driving form part of the UK's employment safety net and how will this function be recreated if demand for professional drivers is significantly reduced?"

The objective for IfDT — an independent UK research and convening body focused on the consequences of autonomous vehicles — is to cast the net a bit wider than simply thinking about whether the software is sufficient to navigate London's wiggly streets and narrow alleys.

The research draws on work with operators trying to bring AVs to London and the regulators responsible for policy. It also relied on more than 30 conversations with councils, operators, industry, unions, charities, insurers, lawyers and universities.

James Hutt, IfDT co-founder, said: "What struck us most in the interviews was how far this reaches: into jobs, public services, the built environment and national resilience. The people who know those areas best are rarely part of the conversation about AVs.

"Our next step is to bring funding and researchers together with the businesses and public bodies that need answers. If you can fund research or offer a place to test it, we would like to hear from you."

In August, Transport for London (TfL) granted private hire licences to autonomous vehicle operators provided that a human driver is present during journeys.

Wayve and Uber have since launched self-driving rides in the UK capital, with 15 electric Ford Mustangs, equipped with the Wayve AI driver, set to pick up passengers with a safety driver on board.

UK company Wayve also demoed a Nissan LEAF, running the Wayve AI Driver using NVIDIA DRIVE Hyperion with self-driving equipment built more closely into the car's design.

IfDT points out that Waymo and Baidu are also preparing services for London, while Uber, Pony.ai and Verne have begun testing in Manchester.

The research organization said UK authorities still need to answer a long list of questions in order to shape long-term infrastructure plans for local areas and prepare for AI's impact on the labor market.

The question of who needs to do this also remains unanswered, as the IfDT asks:

"Which are the key public bodies that should already be anticipating the impact of AVs on their own work and who within those organisations should lead a response?"®

Nanjing Iron & Steel Co., Ltd. (NISCO) has worked with Huawei and ecosystem partners to commercially deploy the world's first steel model powered by swarm intelligence - the YuanYe Steel Large Model (YuanYe) - across multiple scenarios at scale. This helps NISCO transition its core operations from relying on experience to being powered by data and intelligence collaboration.

As a traditional manufacturer navigating the challenges of transformation and cost efficiency, NISCO has built a robust digital and intelligent foundation by adopting industry-leading products including Ascend, Kunpeng, Huawei Cloud Stack, and Pangu models. By embedding AI into every core stage of production - ironmaking, steelmaking, rolling, and quality inspection - NISCO is forging green and high-end AI-powered super steel plants.

Unlike general-purpose monolithic models, YuanYe innovatively fuses multiple AI models with the steel industry's metallurgical mechanism models to create a dual-drive, multi-model cluster architecture. This architecture features a "right brain" for probabilistic computing and a "left brain" for precise and scientific decision-making. Powered by massive amounts of sensing data aggregated from millions of operational technology (OT) data points across 26 business domains, NISCO has achieved real-time digital twin mapping within its intelligent operations center, laying solid data and scenario foundations for YuanYe. This enables the entire steelmaking process to become not only visible, but also capable of precise control and intelligent assessment. Now 100 percent of NISCO's key equipment is connected and digitally controlled and its new product R&D cycles have been shortened by 50 percent, reducing annual costs by over CNY100m (approx. US$15m).

As the model continues to iterate, NISCO had launched 25 business agents by October 2026. These agents cover a wide range of scenarios, from visual quality inspection and process optimization to production scheduling and operations management, setting a new benchmark for digital and intelligent steel transformation.

• In the ironmaking stage, AI makes traditionally invisible blast furnaces transparent. YuanYe leverages blast furnaces' internal sensor data and metallurgical mechanism analysis to perform real-time dynamic simulations of the ironmaking process. Assisted by furnace temperature prediction and health diagnosis models, furnace operators can predict temperatures and identify issues two hours in advance, with over 90 percent accuracy. This shifts furnace operations from reactive post-event remediation to proactive pre-control, increasing prime-grade molten iron yield by over 10 percent while significantly cutting unit energy consumption.

• In the rolling stage, AI drives precise predictions and codifies expert experience to enhance the operational stability of steel rolling and secure leadership in product quality. To tackle the highly dynamic and complex operating conditions during the rolling process, YuanYe's rolling-force prediction model achieves millisecond-level response, narrowing the thickness tolerance to ±0.2 mm and reducing the steel thickness deviation to 1.9 percent to reach a world-leading standard. The rolling-force hit rate has jumped from 88 percent to 98 percent, supporting a steady annual output of over 2.5 million tons of steel plates and increasing the overall yield rate by 0.5 percent. Enhanced dimensional precision and physical-chemical stability allow NISCO's products to meet demanding steel specifications across diverse applications, ranging from mega stadiums and oil pipelines to 10,000-ton cruise ships and LNG storage tanks.

• In quality inspection, the synergy between robots and AI models serves as a discerning "eagle eye", doubling metallographic inspection efficiency. During metallographic inspection for finished steel, the integrated smart metallographic inspection system built on YuanYe has achieved full-process automation, covering sample grinding and polishing, image acquisition, and smart grading. Inspection accuracy has risen to over 95 percent, while per-sample inspection time has been shortened from 10 minutes down to 3-5 minutes. This 80 percent increase in overall inspection efficiency frees human operators to focus on anomaly analysis and high-level decision-making.

• In the energy and environmental control stage, NISCO leverages the YuanYe architecture to develop the S-Link Intelligent Agent for Gas and Flue Gas Sulfur Control. Integrating 602 process parameters alongside multi-modal image and knowledge capabilities, the agent achieves a smart closed loop across source prediction, process optimization, and end-of-pipe compliance. By combining mechanism inversion with time-series forecasting, it issues precise sulfur-risk warnings one hour in advance, achieving a sulfur prediction accuracy of over 95 percent. Operational optimization algorithms then calculate the optimal gas blending scheme within seconds. Finally, game-theoretic models dynamically allocate emission quotas, slashing natural gas co-firing consumption by nearly 50 percent. This has reduced downtime caused by emission exceedance by over 25 percent, and delivered more than CNY600,000 (approx. US$90,000) in annual economic benefits - securing both environmental compliance and drastic cost reduction.

Li Jinyan, General Manager of the Digital and Intelligent Integrated Development Department and Director of the AI Research Institute at NISCO, said: "Digital and intelligent transformation is inherently human-centric. Its purpose is not to replace people, but to unleash human potential. Our goal is to lift employees out of the endless sea of data, empowering them to direct their creative energy toward work of higher value." To date, NISCO has deployed over 300 AI models across its plants, alongside more than 1,000 low-code agents and lightweight applications built independently by frontline employees.

As Cheng Yongfu, Senior Industrial Solution Expert at Huawei's Steel & Non-ferrous BU, noted: "The real value of industrial AI can only be realized when it expands from single-point applications to comprehensive integration with production, operations, and management, which can unlock sustained, scalable empowerment."

Looking ahead, as model deployment expands and computing demands continue to grow, NISCO will team up with Huawei and other partners to explore the construction of dedicated data centers for smart steel plants. Centered on the AI-native three-tier architecture (AI-native OS, AI-native factory, and AI-native enterprise) and powered by the YuanYe model, NISCO is accelerating high-quality steel development, aiming to provide a replicable and scalable benchmark for the digital and intelligent transformation of the global steel and manufacturing industries.

Supplied by Huawei

Welcome to Monday morning, an occasion The Register marks with a new instalment of "Who, Me?" It's our reader-contributed column in which you confess to workplace mistakes and share your escapes from subsequent scrapes.

This week, meet a reader we'll Regomize as "Gordon" who told us he once worked as an engineer at a major semiconductor corporation.

Gordon rose to become manager of an engineering team which worked on a very popular PC chipset. One of the chaps on this team wasn't pulling his weight, so Gordon got the job of telling him to improve, or leave.

"In all fairness, he was a nice guy but not a great engineer and needed to be better," Gordon told The Register.

Not long after that moment of managerial unpleasantness, Gordon's team got the job of doing a full backup of the chipset's "production stepping" - the silicon equivalent of a minor point release. The idea was to create a detailed record that could be used as a restore point.

"This was not a nightly backup, but a purposeful backup to retain a snapshot of the entire cleaned-up design database for future cases where we might need to spin another stepping quickly," Gordon explained.

Gordon told his underperforming colleague to do the backup. "Because I was overworked, I did not think this through," he told The Register.

The at-risk worker quickly proved why he was under the pump by deciding the backup wasn't necessary until the team completed the next stepping - but didn't tell anyone about that plan.

Before that next stepping was done, the employee's past indiscretions caught up with him and he was let go.

The story now fast-forwards six years, to a moment when Gordon learned that a customer wanted to buy the old chipset design so they could make a radiation-hardened version to use on satellites.

The major semiconductor company liked the sound of that, and asked Gordon to find the backup.

"My colleague and I then pieced together what had happened," he wrote. "Being good engineers, we said nothing. We searched through old nightly backup tapes until we found bits and pieces of different steppings of the design database and copied them all to one place, then handed this off to the customer."

To this day, Gordon doesn't know if the customer was able to produce a useful chip from the bits of backup he cobbled together.

"Did they realize they received bits and pieces of two slightly different databases? Did they care? Was the project canceled before anyone realized it?"

"From then on, I was more careful about who I assigned the task of creating restore backups for each design project," he told The Register. "And I kept a record of the emails requesting backups and confirmation that backups had been done."

Have you given a job to the wrong person? Or given clients slightly dodgy data? In either case, click here to send email to Who, Me? We would love to tell your tale on a future Monday. ®

A designer named Marcin Wichary has revealed the story of how he freaked out a decent slab of the world's Firefox-using population with Pac-Man.

Wichary has held senior design roles at Figma, Medium, and Google. As he wrote here, in 2010 Google decided to celebrate the 30th anniversary of Pac-Man by changing its logo at Google.com to a playable version of the game.

The Google of 2010 could still mention its motto of the time, "Don't be evil," without making anyone who heard it guffaw incredulously. The company's frequent changes to the logo on its homepage - called Google Doodles, and which persist to this day - were still a notable part of net culture.

Wichary got the job of building the playable logo, and tweaking Pac-Man to replace its usual maze with a course that instead navigated around the letters GOOGLE.

"It was, frankly, a terrifying proposition," Wichary wrote. "The homepage was then a very important destination, and my job title was only adjacent to 'software engineer.' But I started researching the game and recreating it all in HTML and CSS."

Google wanted a full Pac-Man experience, including sound, "because so many of Pac-Man's sounds are beyond iconic: the starting melody, the waka-wakas, the death jingle."

Wichary began work before Google sought permission from Pac-Man's owner Namco, because the search giant was "planning to wow them with a close-to-finished doodle."

"The project was … going really well. I had fun coding the game and we seemed to have fun playing it. Namco eventually got onboard, too," he wrote.

Google liked the game work so much, higher-ups decided it would be a great idea to have it start automatically if visitors to google.com stayed idle for ten seconds.

The designer liked that idea. "If there was anything scary about it at the time, it was only the thought of my code getting even more exposure - but I wasn't worried about distraction, since the homepage was not a destination page you would ever linger on."

What Google didn't realise was that some people lingered on the home page without ever knowing they were there.

Those people used the Firefox browser with an extension called CoolPreviews that let users hover over a link to see a preview of a web page.

"The extension would quietly start at the same time Firefox was opened," Wichary explained. "And it would immediately, in the background, invisibly and unbeknownst to the user, open a website."

That website was Google's home page, which now featured an auto-playing and version of Pac-Man that CoolPreviews users simply did not expect.

Which explained why, not long after setting the Pac-Man doodle live, Google received plenty of praise, but also reports about computers making strange noises due to CoolPreviews invisibly opening the noisy game.

"I pushed a patch within hours," Wichary wrote, explaining that he added a mute button and a requirement to click something to make the game and its noises start. CoolPreviews couldn't make those clicks, so the phantom Pac-Man sounds went away.

"Ironically, I don't believe we ever heard of normal people being bothered by the autoplay - it seemed that indeed, few people stayed on the homepage for 10 seconds doing absolutely nothing," Wichary wrote.

"The funny thing is that emergency live-patching [the] internet's at the time most valuable homepage was perhaps the scariest part of that day for me, but that completely went over my sleep-deprived head," he added.

Google changed Wichary's title after this incident. His post shows an image describing him as "User experience designer. The Michael Bay of doodles." ®

As debate and legal argument continue over the extent to which developers of large language models are entitled to feed the world's data into their systems without paying for it, India's government has come up with an idea to regulate and secure the flow of information.

India's motive is to create AI models trained on local data, because it thinks offshore model-makers won't be motivated to address the nation's issues or to ensure owners of data stay in control.

The scheme India's Ministry of Electronics and Information Technology (MeitY) thinks can make that happen is called the "Data Empowerment and Protection Architecture (DEPA) for AI Training."

As described by Professor Ajay Kumar Sood, Principal Scientific Adviser to the Government of India, DEPA means "the model travels to the data, rather than the data travelling to the model."

The scheme envisions "Training Data Providers" (TDPs) - examples of which include hospitals, banks, and government agencies - sharing anonymized and encrypted data with a "Data Exchange Platform" (DEP).

Entities called "Training Data Consumers" (TDCs) access the data stored at the DEPs to train their models.

DEPs will operate what MeitY calls "hardware-based Secure Data Environments" in which "data remains encrypted and isolated even during processing." They also enforce rules set by TDPs that describe how TDCs can use data.

MeitY's DEPA framework [PDF] says this scheme means "Differential privacy techniques are systematically applied to prevent the model from learning or revealing individual specific information," adding "All activities inside the secure data environments are immutably logged for audit purposes, ensuring that the process is both secure and transparent."

India's government already operates several data exchanges, but it is unclear how much work remains to make DEPA a reality.

Once DEPA is working, MeitY envisions model-makers will use it to do things like "unlock currently fragmented health data to power breakthroughs in AI-driven diagnostics, drug discovery, and public health management."

The optimism extends to financial services. India is home to literally thousands of banks, and MeitY thinks "DEPA can enable secure sharing of diverse financial data, potentially extending formal credit to 400 million underserved individuals and MSMEs and fueling next-generation fintech innovation."

India's government has long believed that if financial services companies can lend to more people, it will boost economic development.

"Similarly, in Education, DEPA can overcome data interoperability challenges to enable AI-powered personalized learning for millions and create portable skill credentials essential for a future-ready workforce of its 250 million students in a $29 billion EdTech sector," the framework states. "The successful adoption of DEPA Training across these domains is the essential catalyst for realizing India's vision of inclusive growth and responsible AI leadership on the global stage."

While the purpose of the framework is to build Indian AIs for Indians, the document also has half an eye on the nation's international standing.

"DEPA is India's contribution to global science diplomacy, advancing the efforts of institutions such as the World Bank and the United Nations to shape a more inclusive and responsible global AI ecosystem," the framework states.

It also makes several mentions of digital public infrastructure (DPI), an idea pioneered by India that involves government funding creation of tools like the Unified Payments Interface payment systems and the Bharat Taxi zero-commission ride-sharing platform, to ensure Big Tech companies don't dominate the local market.

India also uses its DPI as a diplomatic tool, by giving it away to friendly nations. ®

Google has revealed some of its Asian customers - almost certainly Chinese e-tailers - are buying fewer ads.

Speaking on Alphabet's Q3 earnings call last week, chief business officer Philipp Schindler addressed the slowdown in ad sales growth with the following remark: "In retail, some of our APAC advertisers were impacted by the EU eliminating its de minimis exemption."

That's a reference to the European Union's changes to import duties, which saw the bloc introduce a temporary €3 customs duty for every consignment valued at under €150 imported from outside the EU.

The EU previously operated the de minimis exemption that Schindler referred to, meaning it collected no import duty on almost 5.9 billion low-value items shipped to the continent in 2025 alone. The new regime aims to protect European retailers from cheap imports.

A great many of those items came from Chinese e-tail giants - the likes of Temu, Shein, and Alibaba - who are thought to account for perhaps ten percent of advertising revenue at Google and Meta.

EU policy has therefore contributed to slower growth at Google.

Huawei's automotive efforts appear to have driven it into a cul de sac of controversy.

The Chinese tech giant works with a company called Anhui Jianghuai Automobile Group Co. (JAC) under the Harmony Intelligent Mobility Alliance, which aims to produce intelligent cars.

The flagship model produced by Huawei and JAC is a luxury EV called the Maextro V800, which when tested by Chinese automotive review site Dongchedi produced a rather terrifying fault: the brake pedal bracket broke when drivers exerted maximum force during tests to assess the car's ability to decelerate suddenly at 100 km/h.

The site's drivers were still able to stop the car, but rather more slowly than is optimal.

Within days of the review site posting its findings, reports of its work started to disappear from the internet, according to China Digital Times, a US-based outlet that says it publishes independent news on Chinese affairs.

Even some opinion pieces that called for further assessment of the car's safety disappeared, giving rise to theories that Beijing is displeased with negative coverage of its technology champions and automotive industry.

The USA's Federal Bureau of Investigation (FBI) on Saturday named North Korea's elite cybercrime crew "TraderTraitor" as the source of last week's attack on cryptocurrency exchange Binance.

Cyber-defense company TRM also reached the same conclusion. Its analysts last week declared they have "high confidence" that "on and off-chain intelligence" suggest TraderTraitor did the deed.

"On the evening of October 29, billions of dollars in crypto began leaving wallets belonging to Binance," TRM's analysts wrote. "Within minutes, Binance wallets on the XRP Ledger, Ethereum, Bitcoin, Tron, Solana and four other networks had sent their balances to addresses controlled by a single attacker. A second wave around 23:05 UTC reached Binance wallets on Avalanche and Polygon."

TRM thinks Trader Traitor started its raid 36 hours before the theft. "The attacker's operator wallet received three small withdrawals from Tornado Cash, a service that obscures where crypto comes from, starting the morning of October 28," TRM wrote. "Early on October 29, the operator deployed a first piece of code for the attack and tested it on two wallets. Later that day, it rehearsed the final version on a third wallet, then routed the test funds back to itself. Four hours after that rehearsal, the same code was installed on a Binance hot wallet."

Binance first said the cause of the incident was a single compromised laptop. It has since detected "additional points of breach" but has not offered additional information while it investigates the matter. The crypto exchange continues to reassure users that its platform is safe and that it will make good any losses.

North Korea steals cryptocurrency to fund its government, especially its belligerent defense efforts.

Samsung last week delivered the monster profits it promised early in October.

The Korean giant announced record Q3 revenue of ₩195.3 trillion ($146.2 billion ) and operating profit of ₩107 trillion ($80 billion).

Samsung's DS division, which covers all its chipmaking efforts, earned over 99 percent of that operating profit and $110 billion of the company's revenue.

The company said its foundry business "experienced a notable revenue and earnings improvement on HBM base-die sales growth" and expects it will notch up "double-digit revenue growth and continued earnings improvement on new mobile products, mass production ramp-up of 4nm NPUs, and expanded memory product sales" in its next quarterly report.

"In 2027, the Business will aim for revenue growth led by 4nm and 2nm advanced nodes based on strong AI and HPC demand. It also expects to strengthen its global supply capabilities with the commencement of operations at its new fab in Taylor, Texas," the company said.

That last item is welcome news as the reason Samsung and other memory-makers are making so much money is shortages mean they can inflate prices. More factories coming online may help to ease price pressures.

Dr Rachid Yazami, the scientist who developed the graphite anode that made lithium-ion batteries mainstream, has shown off a new battery design that can achieve 70 percent charge in under four minutes.

In tests of its lithium iron phosphate (LFP) battery cell, Dr Yazami's startup KVI was able to go from 10 percent to 70 percent charge in 3.7 minutes, and to full charge in seven minutes, using "Non-Linear Voltammetry (NLV)" technology that he says "uses a proprietary, stepwise voltage-controlled approach" that "allows the current to respond to the battery's changing electrochemical characteristics" during charging.

"The challenge is not simply to push more electricity into a battery in less time," Dr Yazami said. "It is to understand how the battery responds during charging and use that information to deliver energy more effectively."

Japanese company SkyDrive, which hopes to manufacture electric-powered flying vehicles, has signed up to test the tech. ®

Linus Torvalds has posted the first release candidate for version 7.4 of the Linux kernel and offered his opinion that the project has found a "much healthier model" for using AI.

"This is a *BIG* merge window," he wrote in his weekly state of the kernel update, as it contains almost 20,000 commits compared to the usual ~15,000 that developers submit for a new cut of the kernel.

"I think the reason is the by now obvious one, but I'm actually taking it as a good sign: I'm *much* happier with the flood of AI triggered work going into the merge window than it showing up mainly as extra fixes during the -rc period."

He therefore hopes that the unusually large number of commits "is a sign that the AI fixes are becoming more part of the 'normal development' than a flood of fixes that go into the whole stabilization phase."

"I think that's a much healthier model," he added, before hedging with the observation that "people are expecting 7.4 to be the next LTS [long-term support] release, which is another independent reason why some releases tend larger than others."

"So it may not *just* be the flood of AI patches."

Torvalds also seems not to be completely convinced that all AI submissions are entirely helpful, by describing this release as "Large, and full of fixes for issues found by AI (and more and more the patches have been suggested by AI too, although they often get rewritten)."

AMD GPU register offsets and bit masks make up a third of the changes this time around.

"We've been here before, and I'm sure we'll see it many times again in the future," Torvalds wrote. "Just ignore the noise."

The kernel-watchers at Phoronix rate the major inclusions in Linux 7.4 as improved support for processors from AMD, Apple, and Google. Other additions include some handy tweaks to improve security for virtual machines, support for the Raspberry Pi 10-inch monitor and more work to get Linux running happily on Qualcomm's Snapdragon X2 SoC for PCs. ®

No lesser authority than the MIT Technology Review says that the Emacs editor has turned 50 years old. But then, it is biased — Emacs was invented in MIT's AI lab.

Simson Garfinkel's article is simply called EMACS turns 50, but the sub-heading explains why it matters more:

"How the MIT AI Lab's text editor gave birth to the free software movement."

There are multiple timelines and accounts of the origins of Emacs, and it is not a single monolithic application. The name is an abbreviation for "Editing MACroS", and the original editor for which this set of macros was developed was TECO - "Text Editor and COrrector," written by Dan Murphy in 1962. Murphy died earlier this year at the age of 83.

The inimitable Jamie Zawinski has an encyclopedic Emacs Timeline - as he mentions at the start, his interest being that he long ago maintained a fork of GNU Emacs, originally called Lucid Emacs. It's still around, under its later name XEmacs. There are a lot of forks and re-implementations of Emacs out there, and we are not going to go into all of them, or this article would become a book - one only interesting to programmer-archeologists (to use a term coined by the late great Vernor Vinge).

As for who originally wrote those Editing Macros, though, another history has quotations from several of the people involved: On the Origin of Emacs in 1976. (Note - some of the quotes still contain the | symbols that marked quoted text.)

This credits Lisp boffin Guy L. Steele Jr with the original work in 1975, but in it, Steele says:

"RMS is responsible for the names 'E' and 'EMACS'. RMS still deserves 99% or 99.9% or 99.99% or 99.999% of the credit for taking a package of TECO macros and turning it into the most powerful editor on the planet."

RMS is of course Richard Stallman, and the page also quotes some emails from RMS on October 23 and 31, 1976. Mind you, just a few years after the event, in a 1981 talk now preserved on the GNU website, Stallman said, "I began the development of EMACS in 1974 with an improvement to TECO". Like all the best FOSS, even in its very early stages, it was a collaborative effort - but around late October 1976 is when RMS took over the bulk of the development.

Of the many forks of Emacs over its half a century of development, some are still just about hanging on in there. When we looked up the link for XEmacs, we were surprised to see that there was a new release: XEmacs 21.5.37 "lettuce" appeared just last month. And even the forks have forks: in this case, SXEmacs, last updated in 2020.

Most of the changes in XEmacs and SXEmacs eventually made their way back upstream to GNU Emacs, which is still in active development, although as The Register reported in 2008, RMS is no longer the primary maintainer. GNU Emacs 31.1 appeared in August.

Cards on the metaphorical table: this vulture is not an Emacs user. We have tried. A major influence was one of the Reg FOSS desk's favorite writers, Neal Stephenson, who in 1999 published an essay called In the Beginning… Was the Command Line. (It was later republished as a rather short book, but the text is still online.) In it, he says:

"In the GNU/Linux world there are two major text editing programs: the minimalist vi (known in some implementations as elvis) and the maximalist emacs. I use emacs, which might be thought of as a thermonuclear word processor. It was created by Richard Stallman; enough said. It is written in Lisp, which is the only computer language that is beautiful. It is colossal, and yet it only edits straight ASCII text files, which is to say, no fonts, no boldface, no underlining. In other words, the engineer-hours that, in the case of Microsoft Word, were devoted to features like mail merge, and the ability to embed feature-length motion pictures in corporate memoranda, were, in the case of emacs, focused with maniacal intensity on the deceptively simple-seeming problem of editing text. If you are a professional writer-i.e., if someone else is getting paid to worry about how your words are formatted and printed-emacs outshines all other editing software in approximately the same way that the noonday sun does the stars."

Emacs is 50-plus-year-old software, and it shows. This is a huge and complicated program: decades ago, the joke was that EMACS stood for "Eight Megs And Constantly Swapping". It is not an easy program to learn. Despite lavish documentation and a built-in tutorial, this is software so old that its terminology predates microprocessors and the personal computer.

Its exhaustive manual is 804 pages long, and to pick some trivial examples, its terms for parts of the screen use "frame" to mean what's now called a window, and "window" to mean what's now called a pane. In the section on Keys, it discusses the Control and Meta keys. That's a slight snag: you don't have a Meta key, and neither do the authors of Emacs. It was a key on the now almost legendarily rare Space Cadet Keyboard, of which the Deskthority keyboard wiki says:

"It is known as the holy grail of keyboards."

You can buy a modern replica, the Keymacs keyboard. They are hand-made in the Czech Republic, and depending on exactly what model and specifications you want, they cost from €2,998 for the Lispsaver up to €6,878 for the A624N Space Cadet.

In other words, you can have a keyboard with a real Meta key if you want, but it is going to cost you.

One of the problems with these historical terms, unfamiliar to anyone who wasn't already programming in the 1970s, is that meanings drift. In the context of Emacs, "Meta" means Alt, or Opt on Macs. However, KDE also uses the Meta key - but what the KDE documentation calls "Meta" is the Windows key - or Command on Macs. Like most Linux folks, the GNOME documentation calls it the Super key.

Names are important. They matter. Who's right, the Emacs folks or the KDE folks? Well, neither: nobody has Meta keys any more. We have Ctrl, Alt, and Cmd/Super/Windows.

There have been valiant efforts to modernize Emacs and bring it into line with modern UI and terminology. This vulture tried to learn Aquamacs, which modernized Emacs with a native macOS UI, but sadly it's no longer maintained.

Another, which we enjoyed trying, is ErgoEmacs, which changes the key bindings and rearranges the UI into something much more like the IBM Common User Access arrangement, as used in Windows and most Linux desktops. However, it never really caught on and you need to know your way around Emacs to install it and configure it for your regional keyboard layout.

So complex is it to configure Emacs that there are multiple pre-configured distributions of Emacs, such as Doom Emacs, or the minimalist Emacs Solo. We wish someone did a "CUA-Macs" config, with ErgoEmacs integrated and able to pick up your existing keyboard config.

Come to that, the Gershwin Desktop project that we mentioned along with GhostBSD last year is in active development, and it's based around GNUstep. Maybe its developers could pick up Aquamacs, and adapt it to GNUstep.

There is a long history of such Emacs facelifts. One instance is Borland's Sprint word-processor, one of the selling points of which was that it could emulate the user interfaces of most of the leading MS-DOS word-processors of its day (just before they all converged on IBM CUA). Before Borland acquired it, Sprint was called The Final Word, developed by musical equipment makers Mark of the Unicorn. FinalWord was based on an Emacs clone called MINCE.

Another well-known Emacs distribution is Spacemacs, which adapts Emacs to have a more Vi-like user interface.

Vi is the other seminal Unix text editor, as the Stephenson quote above mentions. As this article on Origins and the Evolution of Vi & Vim notes, it too dates back to 1976. Indeed, that's the reason we chose the year in this story: Tilde, a text editor that doesn't work like it's 1976. One feature of Tilde we enjoy is that it supports use of the mouse even on the Linux console - which is the headline new feature of GNU Emacs 31.1.

Vi is very much alive and well today. GNU Emacs is based on the original source code, but Vi mainly survives as re-implementations. The recreation Stephenson mentions, Elvis, lives on today in the form of nvi.

However, by far the best known is Vim, maintained by Bram Moolenaar until his death in 2023. In the grand FOSS forking tradition, it too has forks, including Neovim, as used in Omarchy, and more recently, the AI-free Vim Classic - a new project led by Drew DeVault, who just launched the similarly LLM-free AlpineBSD. ®

Openreach has issued a stark reminder to customers that the legacy phone network is being officially switched off on January 31, and says the deadline will not be postponed.

The infrastructure arm of BT Group, Britain's biggest telecoms biz, circulated the warning this week in a letter to comms providers that use its network to deliver their services.

It informs them that any customers still on legacy public switched telephone network (PSTN) connections are at risk of disruption when these are retired at the end of January, 2027.

The switch-off could affect businesses and consumers alike. Beyond the plain old telephone, systems that rely on PSTN connectivity include point-of-sale terminals, fire and security alarms, and personal care alarms for elderly and vulnerable people.

The move is part of a wider government scheme to get the UK onto all-digital high-speed networks running on fiber-optic infrastructure, and the legacy phone service - with its dependency on copper cabling - was standing in the way.

Many households have now ditched landline handsets and instead use only mobile phones, yet for those who still want a landline, the changeover means plugging the phone into the back of a broadband router to access a Digital Voice service.

Openreach has been planning this move for the best part of a decade. It has already pushed back the deadline at least once, after the government asked telecoms providers not to force vulnerable customers who use phone-network systems such as TeleCare to switch until fixes are in place.

Nevertheless, public awareness seems low that analog phone services will soon be disconnected. "Openreach and all providers need to coordinate a much more cohesive campaign that reaches customers directly," said telecoms analyst Paolo Pescatore at PP Foresight.

"TV, billboards and newspaper advertising should reinforce a consistent message, supported by targeted letters, calls and door-to-door support for older and vulnerable people who need extra help," he told The Register.

"Customers need to understand what is changing, what action to take, and who will help them. Providers must take responsibility for ensuring people remain safely connected, especially those who rely on their landline or a care alarm."

Caroline Abrahams, charity director at Age UK, agreed.

"With only three months to go, the digital landline switchover is entering a critical final phase. Most people have moved over without major problems, but nobody who relies on their landline or TeleCare should be left behind," she told us.

"Anyone with concerns should contact their provider, but the responsibility cannot rest with customers alone. Providers must not give up on people who have not responded. They need to keep trying to reach them and offer practical, individual support, recognizing that some may be unwell, living with dementia, worried about scams or without anyone nearby to help."

Openreach has a webpage with information regarding the switch-off. This advises consumers to call their service provider as soon as possible to ensure service is not interrupted.

The company previously told The Register that customers without a fiber connection will likely be switched to SOTAP for Analog (Single Order Transitional Access Product), where equipment is installed in the local telephone exchange to allow them to continue to use their old phone line as before.

Openreach has warned that commercial customers delaying migration risk significant disruption and unnecessary costs - leaving it until the last minute could make it harder to secure the requisite engineering support before the deadline hits.

"Businesses and organizations that delay action are making things harder for themselves," said James Lilley, Openreach Director of All-IP.

"The straightforward migrations have largely been completed. What's left are the complex connections that underpin critical operations, requiring careful planning, supplier coordination, and thorough testing. We are urging businesses to act now, audit their remaining analog services, and work with their communications provider on a migration plan today."

Earlier this year, the firm tried to drive home the message by announcing it was stepping up its charges to service providers and companies still operating legacy business lines, who will by now be paying double the rental cost compared to last year. ®

Google on Friday celebrated receiving a poem from its first experimental orbital AI datacenter satellite.

The company's Suncatcher satellite, launched at the start of the month and put into low Earth orbit, has responded to a ground station prompt to "Write a haiku from the perspective of a TPU orbiting Earth, with the wonderful world spinning beneath you."

Travis Beals, senior director of Google's interdisciplinary Paradigms of Intelligence research team, said that with Suncatcher's host and TPU trays powered up and machine learning tests completed, the onboard Gemma 4 model achieved its "Hello World" milestone by sending an inference response from orbit.

"During a ground-station pass, we uplinked our first prompt to our onboard Gemma 4 open model, asking it to generate a haiku about being in space," wrote Beals in a blog post.

And as the satellite passed over Svalbard, Norway, the orbiting AI model responded over the downlink:

Blue marble below,

Data streams in silent grace,

Starlight feeds my core. ✨

Those roughly 20 tokens would cost about $0.000012 if generated through the Google Cloud API, based on Gemma 4 26B's pricing of $0.15 per million input tokens and $0.60 per million output tokens, excluding input costs.

Add in a satellite weighing perhaps 500 kg, based on the mass of SpaceX's Starlink satellites, and a retail launch price of around $7,000 per kg, and you have a very expensive haiku.

According to Google's calculations, if the cost of getting a satellite into low Earth orbit can be brought down to $200/kg, the launch cost amortized over the satellite's useful life (~$810-7,500/kW/year) could be comparable to current electricity spending for terrestrial datacenters in the US (~$570-3,000/kW/year).

Google predicts that SpaceX could achieve that by 2035, more or less, and considers it strategically beneficial to prototype well in advance of that actually happening.

After the haiku, Beals said, "we tested space-themed limericks and telemetry prompts reporting the technical status of the payload systems." Next up: tests for evaluating how TPUs perform under thermal stress and radiation.

Google has already done some laboratory testing of its Trillium Cloud TPU and AMD host server to see how well they can withstand a 67 MeV proton beam.

The system's high-bandwidth memory shows the most sensitivity in terms of the total ionizing dose - it exhibits irregularities after a cumulative dose of 2 krad(Si), which is about 3× Google's minimum requirement for TID survival of 750 rad(Si).

"For typical transformer workloads, the observed SDC [silent data corruption] rate of approximately one event/17 rad, at an estimated orbital dose of 150 rad(Si)/year, translates to a very low failure probability on the order of 1 per 3 million inferences, assuming 1 inference per second," Google researchers said in a paper published earlier this month.

"While this error rate is likely acceptable for inference, the impact of SEEs on training jobs, and the efficacy of system-level mitigations, requires further study."

Indeed, the study will continue. Next year, the plan is to test a pair of interconnected satellites. Rather than lofting large satellites into orbit, Google anticipates sending a constellation of many smaller satellites that communicate with one another. ®