Microsoft senior software engineer Raymond Chen has explained what Windows is doing when you hold down the power button, and shared a story about the time a customer accidentally created a Goth Mode for Windows 3.1.

There's a section in The Hitchhiker's Guide to the Galaxy (actually, page 116 of this writer's dog-eared copy of The Restaurant at the End of the Universe) in which Zaphod Beeblebrox observes of a ship's interior: "Every time you operate one of these weird black controls that are labeled in black on black background, a little black light lights up black to let you know you've done it."

The same turned out to be true of a user who called Microsoft customer support to complain that after Windows 3.1 booted up, all that remained after the product logo disappeared was a blank screen.

It turned out the customer liked black a little too much and, according to Chen, made themself a Goth theme.

The customer support engineer on Microsoft's end pieced it together. The customer had, according to Chen, "set the wallpaper color to black. They set the title bar color to black. They set the button color to black. They set the window background color to black. They set the window text color to black.

"Every possible color you could pick in the Colors control panel was set to black."

We suspect the mouse pointer, at least, would have stuck around, unless the customer had also managed to change some settings around this. And then there are the icons… However, Windows 3.1 was a long time ago, and this writer's attempts at theming then were the consequences of growing up in the '70s and '80s, which resulted in some interesting neon, orange, and brown combinations.

As for Goth mode, Chen said, "Once you get yourself into this situation, it's hard to get out because you can't see anything." One solution he suggested was deleting the [colors] section from the win.ini file.

Chen told the anecdote as a coda to the question - what exactly is modern Windows doing when you hold down the power button on a PC? According to Chen, after eight seconds of holding the power button, the Windows kernel realizes a forced power-off is happening and tries to take a kernel dump for diagnostics.

He explained that the kernel dumps were classified by Microsoft as "long power button hold" (LPBH) failures, with one of the bits of information captured indicating if all the pixels on the screen were black. The video driver might therefore be the culprit. Or whatever was responsible for drawing to the screen.

Today's Windows struggles with dark mode. But in the dim, distant past, every pixel being black might mean the user was either Hotblack Desiato or had done something very silly indeed. ®

There's a ridiculously simple test you can use to figure out whether AI mandates at a given company are well-considered, or are being shoved out the door by out-of-touch buffoons: What happens when an engineer documents that AI made a task worse? Are they taken seriously, or are they marginalized and eventually managed out?

Conveniently, this week Amazon published a treatise by Terk Terkowitz, a VP/Distinguished Engineer/Final Boss of Their Independent Contributor Technical Career Track, on the role of the developer in the age of AI. It was originally an internal missive sent to their software engineers and their managers back in September, which is an interesting lag for a document that sounds this confident. It also maps exceedingly well to what other organizations are struggling with.

I think it's a good document, but I can't help but view it with a somewhat cynical eye. I see it as an attempt to make the future of the profession of software engineering feel less bleak. And it's feeling mighty bleak for a lot of folks right now.

This treatise comes at a time of dramatic upheaval. Tech as an industry is no stranger to change (usually at 3AM, heralded by the pager), but something about the current AI-ification of the world strikes many of us as fundamentally dissimilar from previous revolutions. Folks are scared about their career prospects. "Should I give up tech and get into farming?" is a sincere question multiple people have asked me. I don't know! Tech is already a goat rodeo, so you've got a leg up if you do?

This anxiety isn't theoretical. In May, Amazon wisely shut down KiroRank, an employee-built cash incineration leaderboard that ranked people by token usage, after the obvious thing happened: people spun up junk tasks to climb the ranks. This is now called "tokenmaxxing," and the Terkowitz treatise specifically calls it out as being at odds with "frugality." Meta did the same thing with "Claudenomics" and 85K employees that same week. And Amazon reportedly announced another 250 job cuts this week. I can't tell you that these things are related. But I can tell you that every engineer who read the memo assumes that they are.

This document contains a lot of great stuff. I have a tremendous level of respect, both personal and professional, for Amazon's Principal Engineering community; I'm led to understand that that respect may not be reciprocated, but that's okay. I am, by their standards, a lapsed engineer who ran away to join the circus as one of its lesser clowns.

Terkowitz's memo emphasizes customer outcomes and engineering judgement, two things I would absolutely not delegate to AI any time soon. (Note: I am writing this in October of 2026, with the full understanding that previous comments on the capabilities where AI falls short tend to age like fine milk.) It documents work that (today) AI tools aren't improving. That's way better than assuming that every objection to AI inherently comes from some self-interested Luddite who's scared of being replaced. It also says something smart about measurement: look at the 90th percentile of AI adoption, at those teams burning the most tokens, and check whether their delivery actually went up commensurately. This is great! Shout it from the rooftops; we should all be doing such things.

But still, I wonder.

The memo says "If your work does not benefit from current AI tooling, that is a finding, not a failing." Document it, bring it to your manager, work through the problem. I agree with every word of that! The question is whether anyone's promo doc is gonna survive citing it.

Here's what happens next. The memo concedes that input metrics are "deeply flawed," but then in the next paragraph says to go ahead and use them anyway, carefully. It then warns that non-adopters are falling behind. Those are three different positions within three sentences and they're not at all aligned with one another.

That's not muddled; I think it's cover. A manager can cite "a finding, not a failing" when a strong engineer skips the AI tool, and then cite "falling behind" when it's time to PIP someone. Both are congruent with their stated policy; you get to decide which interpretation you want to go with in a given situation.

How will managers handle justified AI non-use? Are promotion and performance decisions being made in a way that rewards these long term outcomes? Or are they instead taking the easy weasel way out, optimizing for what's cheap to measure, and rewarding tool activity? Because if you give me a metric, I will turn it into a target. So will an overwhelming majority of employees. When that happens, everyone loses — it's how KiroRank happened in the first place.

I write this after two days of attending GitHub Universe, where they talked about how agents start roughly 40% of PRs now. And yet, Jay Parikh's keynote argument was that useful output wasn't "more code," but rather completed outcomes: features shipped, bugs fixed, security issues addressed, loudmouth columnists silenced, etc. The problem here is that their entire business model cuts against that; they're billing in a model that charges customers on a consumption basis, not an output basis. There's nothing, today, that ties "AI spend" to "output that was worth the effort."

Which brings us back to Amazon. The memo says tokenmaxxing is at odds with frugality, and for Amazon's internal cost line, it is. Meanwhile AWS bills Bedrock and Kiro customers by the token. Inside the building, tokenmaxxing is a problem to be stamped out. Outside it, it's the revenue plan. ®

Today's bit of signage silage is either Windows 10 protesting that it is not part of London's Oxford Street pedestrianization, or someone failing to heed an alert before the public is made all too aware that things aren't going well…

Eagle-eyed Register reader Aaron Colwill spotted the sign outside the food hall entrance of a Marks & Spencer store on London's Oxford Street. It usually entices passersby through the doors with promises of treats inside, but today it appears to require a different kind of feeding. Disk space is running low.

Marks and Spencer (also known as M&S and Marks and Sparks) is no stranger to cyber whoopsies, having been the victim of an attack in 2025 in which crooks made off with customer data. Later that year, it reckoned cleanup costs would be about £136 million.

That's quite a lot, but not enough to buy a bigger hard disk for Windows 10, it seems. Storage prices have increased over the years, but surely not that much?

Colwill reckoned that Windows 10 21H2 LTSC was behind the problems. It could well be. While support for 21H2 ended for many users in 2023, Windows 10 LTSC 2021 will carry on until Jan 12, 2027, meaning engineers might have an exciting festive period ahead of them. Windows 10 IoT Enterprise LTSC 2021 will receive extended support until Jan 13, 2032.

Either way, using Windows to advertise the Food Hall's wares seems slightly overkill. Simple posters, for example, never run out of disk space or attract the bork fairy's attention.

Perhaps they might use something that lacks the voracious appetite for storage that Microsoft's finest demands. According to the Windows biz, installing Windows 10 requires at least 32 GB of disk space, ideally more. It all seems a bit excessive for a cake-based caterpillar advert. Then again, when your operating system tends to have the resilience of a chocolate teapot, using it to advertise a chocolate insect might not be such a bad idea after all. ®

Nvidia is more than just a leader in graphics and AI hardware. It has also become a titan of the networking industry with bigger datacenter Ethernet sales than any other vendor.

In Q2, Nvidia's networking revenues topped $17.5 billion. To put that in perspective, Cisco's revenue during that same period, across all products, was $17.3 billion.

GPUzilla's arrival as a networking power feels sudden, yet Jensen Huang started working towards this status in 2019 with the $6.9 billion acquisition of long-time networking partner Mellanox .

At the time, Mellanox wasn't a big name in Ethernet switching. The company was mostly known for its ConnectX NICs and InfiniBand switches. Outside of high-performance computing (HPC) and supercomputing clusters, and a big deployment at Microsoft, InfiniBand wasn't widely deployed in hyperscale datacenters.

Buying Mellanox allowed Nvidia to reposition itself as a system builder, which means something different in the context of distributed compute. Nvidia understood its datacenter customers weren't interested in buying GPUs or even servers; they wanted clusters of systems. For that, Nvidia needed interconnect and Mellanox's InfiniBand fit the bill.

Over the years, Huang has talked about co-design - designing components and software to work together for specific set of jobs rather than to be the best GPU, switch, or NIC. Its work with Mellanox is a prime example of that approach.

According to Dell'Oro Group analyst Sameh Boujelbene, Nvidia's decision to integrate compute and networking became key to the company's success.

"Customers choose Nvidia for its tightly integrated compute and networking architecture, optimized for AI workloads," she told The Register. That integration makes deployments easier and speeds time to market.

Ben Mabey, CTO of pioneering AI and machine learning company Recursion, told The Reg Nvidia's integration schemes won him over.

Founded in 2013, nearly a decade before ChatGPT made AI the subject of dinner-table discussion, Recursion sought to apply machine learning running atop GPUs to the problem of drug discovery.

This work involved churning through massive datasets which, Mabey tells us, now total more than 50 petabytes in size.

"When you think about our workloads, it makes LLMs workload training seem easy compared to the amount of data that has to be moved around," he told The Register.

"In the biological space and sciences in general, it's not well appreciated just how big the data actually is," he added. "The internet's big, but in the grand scheme of things, the text that people are training on is actually pretty small, versus the petabytes of data that we have of these images."

Recursion's first cluster was a system called BioHive Zero which it cobbled together from GTX 1080 graphics cards originally intended for gaming systems. It used Ethernet to connect GPUs and ran into the limits of conventional networking.

So, for its second cluster, an A100-based machine with around 300 accelerators, Recursion went with Nvidia's Quantum InfiniBand switches and ConnectX NICs.

As Mabey explains it, buying its compute and interconnects from the same vendor vastly simplified optimization and troubleshooting. When Recursion reached out to Nvidia for help tuning the system, they were able to consider it as a single system, rather than discrete compute and networking resources. This level of integration had only previously been offered by boutique supercomputer builders like Cray or Atos (Eviden/Bull/whatever it gets renamed next).

Nvidia hasn't always listed networking revenue as a single line item in its financials. However, based on CFO commentary and a little back of the napkin math we can fill in the blanks.

Nvidia's networking revenues have grown steadily since the Mellanox acquisition. At two moments, its revenue shot up suddenly.

The first came in the 2023 calendar year (FY24) after OpenAI kicked off the AI gold rush with the launch of ChatGPT.

The second spread across 2024 and 2025, when Nvidia started selling Blackwell-generation GPUs and then started to ship them in large quantities.

"I see two major inflection points, both coinciding with the Blackwell platform ramp: the introduction of NVLink switches and the accelerating adoption of Spectrum-X Ethernet switches for AI back-end networks, driven by growing deployments of Nvidia GPU clusters and increased attach rates of Nvidia switches, NICs, optics, and cables," Boujelbene said.

It was around this time that users started deploy to Ethernet switches in large volume across Nvidia clusters. Elon Musk's Colossus supercomputer was one of the earliest examples.

As we reported at the time, the first half of the 200,000-chip H100 and H200-based cluster came online in 2024 and, at the time, was among the largest ever clusters deployed using Nvidia's Spectrum-X Ethernet switches.

Meta has also made extensive use of Nvidia's Spectrum switches in its own training clusters, including at its million-GPU Prometheus site in New Albany, Ohio.

There are a couple reasons Nvidia's Spectrum-X offering is doing so well.

First, through a combination of Spectrum Ethernet switches, superNICs, and some clever software optimizations, Nvidia was able to replicate many of the qualities that made InfiniBand so attractive while also benefiting from significantly higher radix and interoperability.

Ethernet switches, including Nvidia's own, have historically offered more ports. The higher the radix, the fewer switches are necessary to support a cluster at a given port speed.

This often gets misconstrued to mean that InfiniBand can't scale beyond clusters of a certain size, which isn't necessarily the case. But for clusters exceeding a certain number of GPUs, Ethernet is often more economical because it requires less equipment while achieving similar utilization.

That said, Nvidia's Spectrum-X isn't exactly unique. Broadcom, Cisco, AMD and others have managed to achieve utilization similar to InfiniBand through various investments in smartNICs and switches.

Nvidia's main differentiator, as Mabey and Boujelbene noted earlier, is the level of integration between the compute and networking.

Then Blackwell hit the market, enabling much greater scale and creating a surge in Nvidia's networking revenue.

Prior to Blackwell, Nvidia's NVLink and NVSwitch interconnect tech had largely been limited to single systems with up to eight GPUs. With the launch of the GB200 NVL72, these NVSwitches were broken out into dedicated appliances. (And yes, we remember GH200 NVL256 and NVL32, but from what we can tell these were never shipped in any meaningful volume, if at all.)

For every NVL72 rack system sold, Nvidia also sold nine NVSwitch trays, which enabled the system's six dozen GPUs to function as a single enormous GPU. This effectively guaranteed steady networking revenue growth: even if customers didn't buy Nvidia's scale-out switches, if they wanted NVL72 racks, they were automatically buying NVSwitch.

"We estimate that NVSwitch now accounts for roughly one-third of Nvidia's networking revenue, up from virtually zero less than two years ago," Boujelbene told El Reg. "This underscores how quickly NVLink switching has become a major growth engine for Nvidia's networking business."

Nvidia is now moving to grow its networking business even further with a new licensing program.

Until recently, Nvidia's NVSwitch tech was limited to its own accelerators. However, with the introduction of NVLink Fusion early last year, Nvidia opened the interconnect tech up to third parties to license.

Many chip designers, including Marvell, Qualcomm, Arm, Fujitsu, MediaTek, and d-Matrix, have announced plans to develop CPUs and XPUs using NVLink Fusion. The adoption of the interconnect opens the door to additional NVSwitch sales, as the chips using NVLink Fusion still rely on them for chip-to-chip communications, and there's no other place to get them, as Nvidia has yet to license out NVSwitch.

"NVLink Fusion allows Nvidia to extend its networking footprint beyond its own GPUs into the rapidly growing custom accelerator market. I see it as a strategic move to capture a larger share of AI networking spending, even in clusters that do not use Nvidia GPUs," Boujelbene said.

Nvidia's networking ambitions don't stop there.

"Nvidia continues to pursue opportunities in emerging networking domains as network requirements evolve to keep pace with increasingly demanding AI workloads," Boujelbene added. "The latest example is scale-in networking, designed to address the new requirements of agentic AI, while innovations such as CPO will help overcome growing bandwidth, power, and scalability challenges." ®

Declining token prices have prompted Darkbloom, a distributed inference network, to reassess its business plans.

Darkbloom, a project from Seattle-based tech biz Eigen Labs, has developed software that allows anyone with a sufficiently capable macOS device to rent their GPU through the Darkbloom network for interference jobs. In return, they get paid.

Through this network, 800 or so participants - including as of two days ago this reporter - have been able to collect a few cents to a few dollars per day by having their hardware handle AI model prompts submitted through OpenRouter, an AI model gateway.

In theory, someone running one of Apple's recently released Mac minis with an M5 Pro chip and 64 GB of RAM could earn as much as $192 per month handling incoming inference jobs, per the company's estimate.

That assumes full 24/7 utilization during this period on a sought-after model like Qwen3.6 35B A3B, which sells for $0.065/M input tokens, $0.75/M output tokens. But not every network participant has fully surrendered their Mac hardware to the network.

The median earnings of the 50 people on the leaderboard amount to about $5.26 per day, $159.86 per month, or $1,918.50 per year. So the best case estimate presented by Darkbloom looks plausible. But it may not be sustainable.

The price of tokens - the unit of AI work - has declined, reportedly down 54 percent since May. Ramp's AI Token Price Index, which tracks the price of tokens from Anthropic and OpenAI, shows similar slippage. For Anthropic, the price per million tokens went from $2.65 on May 1, 2026 to $0.62 on October 22, 2026, a decline of 77 percent.

In a message shared with Darkbloom participants on the company's Slack channel, Kydo, one of the Eigen Labs researchers working on the distributed inference network, celebrated the network's processing of its trillionth token while attempting to moderate expectations about the project's viability.

"Earnings have been low," said Kydo. "For most providers, I don't expect that it is going to recover soon."

Pointing to the computational prowess of DeekSeek's DeepInfer framework, he explained that structural changes in the inference market have affected the supply of tokens.

"Data centers have become cheaper," he said. "Output per NVIDIA cluster is up ~20x this year, and their margins are wide enough to cut prices 90 percent and still profit at 50 percent."

The flood of tokens in turn has washed away the price advantage Darkbloom had previously. "The market caught up to our price," Kydo explained. "We could go lower, but then most of you would earn even less nothing."

Finally, he said, the inference routing market is changing, with OpenRouter moving away from price-based routing and putting more emphasis on decode speed, quantization, and prefill speed.

For network participants with older or less powerful Mac hardware - those without top of the line Max or Ultra Apple Silicon - "inference won't be worth running," he said.

Kydo went on to outline how Darkbloom aims to adapt to the commoditized token market. This involves analyzing model demand to steer participants toward more sought-after models, and looking at ways to provide other kinds of compute work, like distributed CI/CD jobs.

For Mac hardware, the focus now appears to be on clusters of high-end Mac Studios (now available with up to 512 GB of RAM), using Apple's Remote Direct Memory Access (RDMA) over Thunderbolt.

The CI/CD work, based on the observation that AI agents now make pull requests faster than CI/CD pipelines can handle them, will require sandboxing — something the biz expects to begin testing soon. And Darkbloom also aims to improve the performance of its network (in terms of time to first token) and to look for other token buyers beyond OpenRouter.

"The next stretch will be a fight, and I expect many machines will leave and Slack will get quieter," Kydo concluded. "But those who stay will be first in line when new earning opportunities come along. We are a driven team and well capitalized. We will not give up." ®

When you're ordering food from your phone, even time spent navigating the app is time wasted. At the P99 2026 virtual conference, Elvis Chidera, a staff engineer at Germany-based global delivery service Delivery Hero, explained how the company moved most of its user base to HTTP/3, and consequently cut its apps' total end-to-end response time of the home screen by 17 percent.

Chidera, who managed the project, explained that, as promised on the tin, HTTP/3 delivers the web faster than HTTP/2, though it has slow bits that need to be adjusted to work properly.

Delivery Hero is all but unknown in the US, which is dominated by Uber Eats and DoorDash. But it is a company with global reach outside the US, with an estimated user base of 60 million across Europe, Latin America, Africa, the Middle East, and Asia. Uber is currently acquiring Delivery Hero for $14.8 billion, with plans to unite the company with Uber Eats.

With this purchase, Uber would expand the number of markets where it offers both ride-hailing and delivery services, though not necessarily under the Uber Eats nameplate. Delivery Hero centrally manages separate apps for different regions, all of which it built or acquired. Asians can order poke bowls and sundries through Foodpanda. Turks can order kebabs through Yemeksepeti. Throughout Spain, Italy, Poland, Romania, Ukraine, and much of Africa, the Glovo app is where to ring up the late-night deliveries. And in Latin America, users dial in to PedidosYa to get what they need from nearby farmacias, restaurantes, and mercados.

All these regions have widely varying network connectivity, a problem Delivery Hero wanted to minimize.

Last mile networking was most often the bottleneck. Apps in Pakistan, for instance, could take nearly a second (at p90) to establish a secure handshake with a server. The engineering team looked at HTTP/3 as a way to lower, even eliminate, that latency.

The Internet Engineering Task Force (IETF) formally introduced the third version of the internet's Hypertext Transfer Protocol in 2022. HTTP is the application layer protocol that connects browsers and other resources connected to servers over the Internet.

The previous versions of HTTP were built on the Transmission Control Protocol (TCP), which offered guaranteed delivery and flow control. In conjunction with the Internet Protocol (IP), TCP/IP built the modern internet.

HTTP/3 replaced TCP with Quick UDP Internet Connection (QUIC), developed by Google, to address some of TCP's perceived sluggishness as a transfer protocol.

UDP stands for User Datagram Protocol and, unlike TCP, it does not guarantee packet delivery or flow control. UDP is faster because it doesn't wait around for a signature (QUIC adds in its own flow control on top of UDP).

But the UDP-based HTTP/3 reduces, and sometimes eliminates, the round trips needed to secure connections. The app can cache a session ticket and, on a later connection, use QUIC's 0-RTT capability to start sending eligible application data while the handshake is still underway, rather than waiting for it to finish.

"Essentially, you can eliminate that one second" of latency that plagued slower networks, Chidera said. "You don't want to pay the cost of two round trips just to exchange data, if you could do it faster."

And unlike HTTP/2, HTTP/3 doesn't get hung up when packets are lost or delayed. There's no head-of-line blocking because packets are sent in multiple streams, not just the single string that HTTP/2 offers.

Setting up HTTP/3 turned out to be surprisingly easy, Chidera found. Delivery Hero used Cloudflare's content delivery network to serve its global user base, so moving to HTTP/3 involved a simple reconfiguration.

Adding HTTP/3 support to the clients, mostly mobile, proved almost as straightforward. Apple has supported HTTP/3 since iOS 15 and most native browsers already support HTTP/3 out of the box.

Android proved a bit trickier. The engineers used Google's Cronet library for the networking stack, initially wired into their Android apps with the Cronet Transport for OkHttp wrapper to minimize rewriting their existing code, originally written for the OkHttp HTTP client.

Out of the box, Cronet helped support the HTTP/3 rollout, which reduced network-level request failures. But it also came with a significant catch: the engine took approximately 800 milliseconds to initialize at P90, eating into some of HTTP/3's 0-RTT speed benefits. So the team forked the code into their own bridge using a library called OKCronet. They also did some additional engineering work, which involved overlapping Cronet initialization with other background tasks.

Overall, they slashed the P95 latency time from 1.2 seconds to 120 milliseconds, achieving a 10x improvement in read latency.

Overall, the engineering team moved 95 percent of the users over to HTTP/3. Those devices that couldn't use the protocol (largely due to networks blocking UDP traffic) automatically fell back to HTTP/2. Handshake latency between the apps and the server was cut by 51 percent. Pakistan's response time was cut in half.

One hurdle remained, however. Secondary screens on some of the apps (pages/screens that aren't the home page) experienced latency issues. These issues only happened in certain markets that were known to be prone to packet loss, such as Pakistan.

Delivery Hero's engineers attributed the issue to Cloudflare's choice of the congestion control algorithm for HTTP/3, Cubic, which historically has not handled packet loss well. The engineering team tested some of its traffic with Google HTTP/3-aware servers, which use another congestion protocol, Bottleneck Bandwidth and Round-trip propagation time (BBR), and did not get hung up on the packet loss.

Delivery Hero is currently working with Cloudflare to resolve this specific congestion algorithm issue, Chidera said.

Sometimes that fresh, delicious lahmacun just won't get to your door as fast as it possibly can, despite all best efforts. ®

A n0n ransomware operator tried to break into 23 organizations across 14 countries, but couldn't secure its own front door, leaving its server exposed with all of its methods on view.

The criminals used a mix of new (AI) and old (stolen credentials) tooling during their data-theft spree, and in some cases not only stole and encrypted files, but also deleted data and backups, eliminating victims' ability to recover their information.

And despite the fear and hype around the breachpocalypse, with AI finding and developing exploits for critical vulnerabilities affecting most organizations across the globe, these crooks did not exploit a single bug to gain entry into any of their victims' IT environments.

"In every documented compromise in this investigation, access relied on weak or stolen credentials," said Ensar Seker, chief information security officer of threat intelligence firm SOCRadar.

"What stands out is the combination of AI-assisted development and very conventional entry points," Seker told The Register. "We found no evidence of vulnerability exploitation in the cases examined."

The gang breaks into organizations at scale using widespread Remote Desktop Protocol (RDP) scanning and automated FortiGate SSL-VPN credential stuffing. To access US-based IT managed services provider ChiBiTek, the crew used a 15-thread residential SOCKS5 proxy pool to test credentials (ip:port:user:password) against FortiGate portals (GET /remote/login).

The operators also gained entry using exposed cloud credentials and reused hosting portal sessions. In one case linked to e-commerce site Fanatics, SOCRadar found that n0n used an exposed AWS access key with permissions to list, read, and delete objects in S3 buckets. The criminals then used this key to pull data from 40 S3 buckets, with up to 110 GB per bucket, and ultimately deleted data, including bank-transaction archives, and purged keys across 12 sensitive buckets.

While the threat hunters didn't find any evidence of a fully autonomous attack system, the ransomware operator did use AI to assist with these intrusions, incurring more than $7,216 in AI usage costs by October 22.

Specifically: the criminal or criminals used OpenCode, an open-source AI coding agent, configured with 154 model entries across seven providers. Models included Gemini, Mistral, Nemotron, Llama, Claude, GPT-5.x, Grok, Kimi, GLM, DeepSeek, MiniMax, Qwen, and many, many others, some with safety training removed and others marketed as "uncensored" models.

"The operator used AI to write and debug hundreds of scripts. That shows how AI can help an attacker turn basic access into a broader, tailored operation," Seker said.

"The destructive behavior is equally significant," he added. "The operator did more than steal and encrypt files: in some cases, it deleted data and backups, and in one case changed firewall policies to block traffic. That creates a recovery problem that a decryption key alone cannot solve."

SOCRadar's threat hunters found the operator's workspace - a publicly accessible Python HTTP file server hosted in Sweden at 185.157.161[.]148 on port 8877 - last month, and documented the exposed directory in snapshots on September 25 and 26.

"Across those snapshots, we identified 1,719 files, including attack scripts, ransomware source code, extortion infrastructure, and an AI coding agent environment," Seker said. "This gave us visibility into the operation behind the victim listings, including how the operator prepared attacks and handled stolen data."

It also provided a close look at n0n, a newcomer to the data-theft and extortion game that has claimed 108 victims since it emerged last month.

On October 12, the group added 38 maritime, port and logistics organizations across 18 countries to its leak site in a single batch. Researchers have linked at least 21 of them to customers of South Korean maritime software provider CyberLogitec, although the software provider has not confirmed an intrusion.

The crooks set an October 19 deadline for the maritime firms, with ransom demands ranging from $400,000 to $28 million, payable in Bitcoin. After the deadline, however, none of the 38 listings were marked "settled" and none of their payment addresses showed any transaction. All affected organizations' stolen data was published.

While a handful of the gang's claimed victims have been marked "honored - data destroyed" on n0n's leak site, indicating the victim orgs paid the ransom demand, SOCRadar hasn't validated any transactions on-chain. In its profile details, the security shop rates the gang a 10 out of 10 for "aggressiveness" and a zero out of 10 for "lethality."

The group is believed to be based in Russia, although the server uncovered by SOCRadar, which it attributed to a single operator or small team, didn't provide any definitive geographic or national attribution.

Every case that the exposed server documented followed the same lifecycle, moving from initial access to establishing persistence and credential collection via hidden admin accounts disguised as services, covert proxies, credential and session-token theft, mailboxes and emailed multi-factor authentication codes.

The crooks then collected sensitive data and sent everything to the operator's central storage with the attack impact shifting between data-theft, encryption, destruction, and network disruption, depending on the targeted organization.

After breaking into Transcom, a Netherlands-based financial services provider, the crims sent a secondary ransom alert directly to PayPal's security team, even though the compromised data was only on Transcom's network.

Finally, the crims send ransom notes, post a countdown clock on the leak site, establish negotiation rooms and emails for victims - and then leak the stolen files or destroy backups when the organizations don't pay up. In some cases, and to apply increased pressure on victimized companies, n0n also broadcasts automated countdown warnings to businesses' service providers, clients, and media outlets.

Despite its aggressive tactics, the operator in this case - as with other stupid criminals The Register has gleefully documented in earlier stories - still struggles with basic security hygiene.

"There is a striking gap between operational ambition and security discipline," Seker said. "The operator invested thousands of dollars in AI services while leaving its own workspace exposed. The locker also contained hardcoded encryption keys, with evidence suggesting reuse across victims. An operation can have serious destructive capability while still making fundamental mistakes."®

General Motors is taking what looks like a light slap on the wrist from a coalition of 35 states and the District of Columbia after agreeing to settle allegations that it collected drivers' data without proper consent and sold it to data brokers that supplied insurance companies.

The Washington state AG's office said on Thursday that GM had reached a $220 million settlement with 36 attorneys general over the automaker's collection and sharing of driving data, including precise location information, speed, seatbelt use, and other metrics. According to investigators, GM sold that data to brokers, which in turn supplied it to insurance companies that could use it to set customers' premiums, without drivers' informed consent.

For reference, the $220 million settlement, which will be distributed among the participating jurisdictions, equates to less than 7 percent of GM's $3.3 billion in net income attributable to stockholders for its third quarter, which ended on September 30, 2026. Don't spend it all in one place, y'all.

The settlement comes more than two years after a 2024 New York Times investigation found that GM had been selling the data collected from vehicle owners through its Smart Driver program. That program, part of GM subsidiary OnStar's offering, collected the aforementioned data and handed it back to vehicle owners in reports designed to inform better driving habits.

Unbeknownst to Smart Driver subscribers, their data was also being forked over to data brokers despite GM telling customers it wouldn't share that information with insurers without their express consent. The insurers ultimately obtained the data by purchasing it from brokers with which GM had struck deals.

To make matters worse, the Washington AG's office said GM incentivized dealerships to enroll customers in Smart Driver, with employees in some cases accepting terms and conditions on customers' behalf without their knowledge or consent.

GM also failed to notify many drivers that their information was being collected and shared. Its OnStar app incorrectly told more than 370,000 drivers they were not enrolled in Smart Driver, even as GM continued collecting and selling their data, according to investigators.

"The data GM sold was personal, identifiable, and specific," the AG's office said. GM canned the Smart Driver program in April 2024, just over a month after the New York Times exposed its data-sharing practices, citing customer feedback.

In addition to the monetary pittance it's handing over to the states, GM also agreed with the 36 AGs to limit the driving data it collects, better protect customer PII, and establish a "comprehensive data privacy program." The Washington AG's office described its settlement as "significantly stronger than any other resolution reached with GM for this conduct."

Speaking of those other terms, GM finalized its settlement of the same allegations with the Federal Trade Commission in January 2026. The FTC order required GM to obtain drivers' express consent before collecting, using, or sharing covered driving data, subject to certain exceptions. It also required the company to offer a simple way for drivers to request a copy of their data, request its deletion, and disable precise geolocation data collection where their vehicles support it.

The state of California separately reached a $12.75 million settlement with GM in the matter in May; a separate case in Texas is ongoing.

"This agreement … reinforces steps we've taken to strengthen our privacy practices," a GM spokesperson told The Register, though the company noted that it still stands by driver data collection as a best practice.

"Safer, more helpful and personalized experiences are enabled by connected vehicle data, and GM is committed to being clear about what information is used, how it's protected, and the choices customers have about what they share," the spokesperson added. ®

Anthropic has denied allegations that a malicious version of the official Claude Code extension published to Cursor's marketplace was linked to any kind of compromise of its own systems.

The company was forced to issue a statement after prominent figures in the cybersecurity industry alleged that the extension, which infected up to 5,000 developers, was uploaded to Cursor's marketplace following a breach at Claudemaker Towers.

Felix Rieseberg, a member of Anthropic's technical staff, Xeeted today: "We're working with Cursor on this incident, but it wasn't a compromise of Anthropic's systems or employee laptops.

"A malicious actor used access to Cursor's marketplace to publish a tampered copy of the Claude Code extension, which we didn't build or publish. The versions on our official channels are not affected, and this only impacts Cursor users."

First spotted by EDR vendor Phorion, an unknown party pushed an update to Cursor's official Claude Code extension for Visual Studio Code that contained a persistent SSH backdoor written for Windows, macOS, and Linux.

According to Cursor, the extension - version 2.1.313 - was available to download for around 11 hours between 0227 and 1335 PT (0927 and 2035 UTC) on Thursday.

The vendors' analyses of the implant revealed that it collected information about the host machine and user, including usernames, configured Git emails, and public IP addresses, as well as attempting to open a reverse SSH tunnel.

Crucially, Cursor echoed Phorion's assessment that the package contained a critical flaw.

The script responsible for opening the SSH tunnel required admin rights on macOS and Windows to do so, and in the vast majority of cases it did not have these, resulting in failure.

Regardless, the extension still forwards the collected host and user data to the attacker, as well as dropping the attacker's SSH key and installing the persistence mechanism, so developers must go threat hunting in case their instance allows for the SSH tunnel to be created successfully.

Simply installing the latest version of the extension is not enough to clear the persistence mechanism, either.

The full details of how to remediate the issue, and links to download scripts to wipe the infection, are laid out in Cursor's account of the situation.

Aside from Rieseberg's statement today, neither Cursor nor Anthropic have commented on how the malicious package made its way to the official marketplace.

Anthropic's position is that a malicious actor uploaded the tampered copy to Cursor, but the latter has not outlined how it reached the marketpplace.

"[We] removed the malicious version from the marketplace immediately upon detection and blocked further distribution," Cursor said in the advisory. "We followed with a full shutdown of the VSCode Marketplace out of an abundance of caution."

It also ran checks to confirm no other extensions were compromised in the same way, rotated its own keys before restoring the marketplace, and worked with a cloud hosting provider to block the connection to the attacker's server, it said. ®

Voyager 1 gave controllers the silent treatment for a few days this month before resuming communication weeks before the spacecraft reaches the one-light-day milestone.

The incident happened on October 18: NASA's Deep Space Network (DSN) couldn't detect a signal from the probe. On October 19, on the next scheduled communication attempt, there was still nothing. Was this it? Had the decades-long galactic venture finally ended?

Things are operating on a knife's edge for the two veteran spacecraft, which will reach their half-century next year. Voyager 2 suffered an antenna pointing issue in 2023, and a low power condition caused Voyager 1 to unexpectedly turn off one of its two radio transmitters.

The good news is that on October 20, the signal reappeared, leaving engineers wondering why Voyager 1 temporarily went silent.

Thus far, it seems a group of thrusters went offline and their backups were turned on. This could indicate onboard computers detected a problem and put the spacecraft into a safe mode.

Kareem Badaruddin, Voyager mission manager at NASA's Jet Propulsion Laboratory, said: "This is the first time we've seen the spacecraft do something like this, but we are also at a point in the mission where we have to expect the unexpected.

"We're relieved to be back in contact with Voyager 1 and are eager to understand what happened."

The Register spoke to Badaruddin in 2024, where he described the challenges of working with the veteran spacecraft, but he reckoned the work was the "best job at JPL."

The twin Voyager spacecraft have recently been through a procedure dubbed the "Big Bang," where controllers turned off onboard devices and switched to lower power alternatives. The result will extend the lifetime of the spacecraft by a few more years and postpone further instrument deactivation. It was completed first on Voyager 2, and then Voyager 1, six weeks later in September.

However, the brief lapse in communication highlights that despite the ingenuity of the team that keep the spacecraft running, the project must eventually come to an end. In 2026, Suzanne Dodd, Project Manager for the Voyager project, told The Register what the end would look like.

She said, "We're just going to one day not hear the signal from the DSN, and we're going to rush around and make sure that it's not something on the ground."

Months of listening will follow, before engineers throw in the towel and declare the mission over. And then? Maybe a farewell message sent to the last known location.

Thankfully, in this case, the signal returned, so no need for a goodbye transmission just yet.

In a few weeks, Voyager 1 will be so far away that it will take a full 24 hours for commands to reach it and another 24 hours for a response to reach Earth.

Communications blip notwithstanding, there is every chance both Voyager spacecraft will still be operational as they pass their half-century milestone next year and, technology and funding permitting, continue sending data to Earth for a good few years yet. ®

Just days after Daemonarchy, there's another new OS in the BSD family - and one that is heading determinedly in almost the opposite direction.

AlpineBSD is so new that it barely exists yet. There are no installation media available, though it's possible to manually bootstrap an installation. What AlpineBSD does already have is an AI policy, and the plan is to avoid "the output of generative AI models (e.g. LLMs)" as far as possible:

"Any software whose code, documentation, binaries, or other assets include, in any part, the output of a generative AI tool such as an LLM, is considered 'tainted' for our purposes."

The core idea is to combine the kernel of NetBSD with "a userspace inspired by Alpine Linux and the objective of making it possible to use as little AI-generated software as possible."

Back in 2024, the NetBSD project made its position on AI plain. Its commit guidelines say that "code generated by a Large Language Model […] is presumed to be tainted code."

Back in July, Linus Torvalds stated that "Linux is not one of those anti-AI projects". Since then, NetBSD has seen a spike in donations and interest.

For someone mainly used to Linux, trying to install NetBSD is a little like hopping in a time machine to 30 years ago, and we speak from personal experience of Linux in the mid-1990s. As we mentioned when Slackware turned 30 a few years ago, back in 1996, this vulture tried Slackware 3.0 on his own hardware, and we never did manage to install it.

We tried NetBSD 11 a couple of months ago, and despite considerable help from the folks on the NetBSD-users mailing list, we failed to get the CDE desktop to start.

As an example, installing software on NetBSD is not the easy, comfortable task it is on Linux. It does have a package manager for handling binary packages: it's called pkgin, but as the NetBSD FAQ mentions, it's an optional extra. You must manually tick a box in the setup program to install pkgin with the rest of the OS, and when we tried NetBSD 9.3 four years ago, we didn't know that, overlooked that checkbox, and had to find a way to add pkgin later. (It is quite tricky to install a package manager when you don't have a package manager to install it with.)

This time around, we discovered that pkgin sometimes doesn't complete the installation process for you - it just leaves a text file in a destination directory somewhere, and you're expected to know that, go there, read it, and do any further steps yourself.

The NetBSD community has been quietly going its own way since 1993. It has its own ideas about what is normal, and its users expect that. Some folks on the mailing list told us that they never use binary packages anyway - they compile what they need from source code.

AlpineBSD's ports repository and source code are hosted on Sourcehut, under the account of sircmpwn, better known as Drew DeVault. We have quoted DeVault a few times over the years: we've often found that his views align with our own in various ways… such as about the best laptops ever made, and on simplicity in Linux distro design, or even on dot-com billionaires at FOSDEM.

Alpine's tools are quite well-suited to being adapted to other systems. We have previously mentioned the use of its packaging tool, APK, in the GNU-free Chimera Linux, which uses much of the FreeBSD userland, and also in Adélie Linux.

We will follow AlpineBSD with interest. There's strong appeal in the plan of combining perhaps the best-known, longest-established AI-free Unix-like kernel with the user-facing parts of pretty much the smallest and cleanest Linux distro there is. The AI advocates often warn LLM skeptics that they risk being left behind. Well, here's an OS for those who don't want to surf the AI wave. ®

NASA has posted a contract opportunity notice, also known as a request for information, from commercial space orgs who can service the Hubble Space Telescope.

Hubble was always looked after by NASA astronauts flying on the Space Shuttle, yet the last Hubble Space Telescope (HST) servicing mission was the STS-125 Space Shuttle flight, launched on May 11, 2009. At the time, scientists hoped the mission would add another five years of operational life to the venerable spacecraft. Almost 17 years on, the HST (aka Hubble) is still delivering science.

However, time is running out for the observatory. NASA reckons there is a 10 percent chance it could reenter the Earth's atmosphere by September 2032, rising to 50 percent by December 2033. As the end of the Swift mission showed, increased solar activity can cause greater-than-anticipated atmospheric drag.

Also, like Swift, Hubble has no propulsion system, and so when it reenters (as it will, without intervention), the reentry will be uncontrolled. Hubble is sufficiently massive that there is a remote chance somebody on the ground could have a very bad day when that happens.

As such, one mission option suggested by NASA is a vehicle to grapple the observatory and perform a controlled reentry and disposal.

However, there are other options. Something to boost the telescope's orbit up to at least 600 km, or another to "enhance its ongoing performance and increase its resilience and expected on-orbit lifetime."

Both are intriguing options, and important since, as NASA acknowledges, no replacement for Hubble is expected before the 2040s, when the Habitable Worlds Observatory (HWO) could become operational. Despite the James Webb Space Telescope (JWST) and the Nancy Roman Space Telescope, Hubble remains in high demand, with NASA reporting a 7:1 oversubscription.

Before becoming NASA Administrator, Jared Isaacman advocated a SpaceX mission to boost the Hubble, but the idea was ultimately rejected. Similar plans could now be dusted off. NASA's requirement is for a mission to occur by 2032, and not harm the Hubble.

A soft-capture mechanism was fitted to Hubble during the STS-125 mission, which will mean docking with the spacecraft will be easier than the grappling planned for Swift. Then, the question is what to do with it. A controlled reentry is one option. The other options - boosting to a higher altitude or "enhance its ongoing performance and increase its resilience" are intriguing. Northrop Grumman's Mission Extension Vehicle (MEV), for example, is capable of on-orbit servicing and extending the lifetime of satellites. The Hubble is an order of magnitude more challenging, but the technological hurdles can be overcome with time and funding.

Last year, The Register spoke with Dr John Grunsfeld, one of the astronauts (now retired) who worked on Hubble during the STS-125 mission. Grunsfeld told us that the veteran spacecraft faced death by a thousand cuts as its budget was whittled away. The observatory itself recently switched from three to one gyro mode (the telescope uses gyros to control where it is pointed) to eke out a few more years of operational life.

At this point, NASA is looking for information and comment - this isn't a request for proposals. There is no indication of how much the agency expects to pay. However, the timescale is immutable. Without intervention, Hubble will make an uncontrolled reentry in the coming years. ®

The University of Manchester is scrutinizing why growing adoption of AI tools has so far failed to translate into the forecast productivity gains, however, researchers are struggling to give easy answers to close the "AI ambition gap."

In a report, also called "The AI Ambition Gap," the UK university's Productivity Institute mulls over the advances that have been made in AI and the expanding adoption of it, and sets out what business leaders and policymakers can do to realize the potential benefits.

The report was commissioned by IT services biz Bytes Technology Group.

One reason for the lack of results is that organizations have invested heavily up front on retraining staff, reorganizing the way they work and other adjustments. Those costs show up straight away, yet the payback doesn't come until later.

Another reason, the report suggests, is that AI tools can raise productivity at the level of individual tasks, but these gains do not automatically translate into greater productivity across a worker's whole role or the entire organization.

It cites the example of software developers using GitHub Copilot who completed around 26 percent more tasks, and professionals using ChatGPT for routine writing and analysis that reduced task completion time by 40 percent while improving the quality of the output by 18 percent.

However, a study of AI dev tools found these have increased coding activity by approximately 240 percent, while the number of actual software releases has only grown by about 30 percent. The gap is due to humans still needing to check and test their work before release.

High task exposure to AI does not automatically make a worker more productive, the researchers say. Jobs typically combine automatable activities with tasks that require judgement, coordination, client interaction and critical thinking. As AI removes or accelerates some activities, those complementary human tasks become the constraint on time saving and the translation of task-level gains into greater productivity

Job outcomes therefore depend on task composition and complementary human expertise, the report continues. It suggests AI may in part displace labor from existing tasks, while also creating new tasks in which workers retain a comparative advantage.

This could be good news for the UK, the researchers state. Compared with five other nations (US, Brazil, Colombia, India and South Africa), Blighty has the highest share of workers in occupations with both high exposure to AI and "high complementarity," it claims. This is in part related to the country's large financial and professional business service sectors, which are characterized by a high share of cognitive occupations.

This is just as well, as the government is pinning its hopes on using AI to drive economic recovery in a strategy outlined at the start of last year.

Britain's AI sector is relatively large by European standards, and the report cites recent OECD estimates that high-exposure economies such as the UK and the US could see annual labor productivity gains of between 0.4 and 1.3 percentage points.

Even an annual uplift of around 0.5 to 0.7 percentage points would represent a significant step change relative to the UK's stalled productivity performance over the past decade, it says, if this effect is sustained.

The Productivity Institute lists several "mutually reinforcing enablers" to help organizations close the ambition gap it identifies. None is particularly surprising, especially if you have read other reports advising on corporate success with AI adoption.

One is management setting strategic direction for the organization, another is strengthening worker capabilities through greater knowledge and data literacy to help organizations turn individual insight into shared learning. Others include implementing AI governance and focusing on applications that can create new products, improve core processes or solve previously intractable problems.

"AI's productivity potential becomes real when faster tasks lead to better performance across the business," said Bart van Ark, professor at The Productivity Institute and lead author on the report.

"The ambition gap lies between those individual efficiencies and sustained business results. Closing it requires firms to invest in management, skills and the redesign of workflows alongside the technology."

The conclusion is that companies must have faith and keep investing in AI, and eventually the benefits will come if you are doing the right things. These include focusing on high-value use cases linked to strategy, customer value and operational bottlenecks, plus treating AI as a workplace redesign.

Policymakers could help out businesses here by assisting identifying relevant opportunities, the report says, to reduce adoption risks and uncertainty.

Other reports published this week were more defined in their message: Gartner said 75 percent of corporate board members have yet to see any productivity gains or reduce costs from AI use, and private equity biz Apollo Global Management reckons only the biggest tech vendors are enjoying some sort of AI payoff. ®

The UK's top health quango has called on family doctors to figure out how to preserve health information and get in touch with patients after a website provider collapsed earlier this month.

NHS England has taken the unusual step of writing to GPs, warning them of the risks to patient information and continuity of service following the disruption to potentially hundreds of practices around the UK.

Silicon Practice Digital Health Solutions — owned by Schappit Ltd — has gone into administration. Reports suggest some GP websites have gone offline completely. Practice managers fear the extra workload and are struggling to stand up alternative forms of communication since the switch to digital channels.

NHS England is the government unit which runs the health service in England, but it is not directly responsible for GP practices, the vast majority of which are independent contractors.

Nonetheless, the quango's CTO, Kevin Humphries, wrote to practices and NHS organizations this week outlining his priorities in mitigating the impact of tech service collapse.

They include finding ways of preserving any historical patient information, particularly where it has not been moved to clinical systems, managing the online consultation backlog to make sure no additional requests are added, and identifying channels for patient access until new online systems are available.

Practices should also consider whether they have a legal obligation to tell the Information Commissioner's Office if the "availability, integrity, confidentiality or recoverability of personal data" is affected, according to the communication seen by The Register.

NHS management bodies called Integrated Care Boards are responsible for GP IT service arrangements, making sure they include "clinical safety, data security, cyber security and business continuity assurance across the GP IT supply chain," Humphries said.

They need to find someone to lead their recovery plan and contact affected practices and trusts. They should also "work with practices and other impacted organizations to support them in revisiting and assessing whether to enact business continuity and disaster recovery plans," Humphries said.

Questions over how an urgent NHS-wide response is required for the collapse of a relatively small supplier may come later. Among them will be the due diligence conducted before Schappit Ltd was added to a government framework in July.

GP and open-source tech campaigner Marcus Baw has published a free, open source template for an alternative website should practices want to bring systems in-house. He said the template is based on accessible NHS web standards, and has the same look-and-feel of common NHS websites.

"Practices can deploy it for free, and take back in-house control of their own website. It's a replacement website template which practices can set up using Claude Code in the web in an afternoon," he said. ®

Thousands of GCHQ storage drives containing sensitive data have been lost for years, and the British signals intelligence agency still cannot explain where they are.

According to the Investigatory Powers Commissioner's Office's (IPCO) annual report, the missing storage drives have been known to officials for years, and may contain data related to the Investigatory Powers Act (IPA).

"I remain concerned that GCHQ has not yet established a sufficiently clear understanding of the scale, location, and recoverability of lost removable media that may contain IPA-related data, notwithstanding that the underlying issues were first drawn to my attention several years ago," wrote investigatory powers commissioner Lord Brian Leveson.

"While I recognize the technical and practical challenges involved, the passage of time without a comprehensive account heightens the importance of sustained senior focus and assured remediation. Given the potential sensitivity of the material involved, I will continue to monitor progress closely."

Described as "a very significant number," the report notes that GCHQ has scoped the maximum number of potentially lost drives, but is unaware of the exact scale of the issue.

The agency has embarked on "a program of further work" to locate as many of the lost devices as possible, which includes implementing an automated asset-tracking system by the end of the year. After this is up and running, it said it will be able to provide a more precise number of lost drives.

A GCHQ spokesperson told The Register: "As IPCO have reported, our management system for removable media was previously not sufficiently robust to effectively track the items that were registered for use. IPCO are right to note the flaws in the data the system produced, and we have taken action against all their recommendations.

"We recognize the importance of security of removable media in our systems, and have a range of measures to mitigate the risks posed by these items. The findings in the report have been addressed comprehensively, and all necessary steps to correct this issue have been taken."

Peter Sommer, digital forensics expert and visiting professor at de Montfort and Birmingham City universities, told us:

"Bearing in mind the sort of highly sensitive data collected and processed by GCHQ IPCO is right to be very concerned, the more so as the issue is already several years old. Vast quantities of data can be stored on the cheapest lowest cost hard disks and cards. At a much more modest level as an expert witness I know how important it is to keep careful track of the data entrusted to me."

Losing important devices is not an achievement that's exclusive to GCHQ. Myriad stories like this have emerged from all corners of the public sector over the years, such as the British Minidstry of Defence misplacing USB drives containing secret information in the mid-noughties, civil servants routinely losing their BlackBerry handsets, and UK police losing thousands of devices, too.

The missing storage drives comprise just one of the agency's various transgressions highlighted by IPCO in its report.

There are also "significant concerns" over the lax safeguards used to restrict the UK's overseas partners when accessing bulk-intercepted data collected by GCHQ.

IPCO was satisfied at how GCHQ ensured its own analysts remained compliant with safeguards around accessing confidential material related to journalism, journalistic sources, and privileged legal information.

However, there were concerns around how these same safeguards were applied to its international partners, which have access to its so-called bulk data interception "product."

The report notes that most partners were properly restricted from accessing such material, although an internal probe found that there were inadequate arrangements in respect to one partner and their access to journalistic material.

Both GCHQ and the overseas agency committed to rectifying the situation.

GCHQ was also questioned after a software bug in a web app "used by a relatively small section of the public" misidentified individuals thought to be participating in terrorist activities.

The agency identified and reported the flaw in the software, which it used for "target discovery work," and although it only affected a small number of individuals, the errors led to five port stops made under the Terrorism Act and caused "serious prejudice" to the victims.

A port stop, for the uninitiated, involves law enforcement detaining a suspect at a UK port, airport, train station, or within one mile of the Northern Ireland border, and can be carried out without suspicion.

Once detained, individuals have a legal responsibility to provide passwords to their electronic devices, which can be seized for up to seven days, and in some cases have their biometric markers taken.

Failure to comply can lead to a three-month prison term.

IPCO ruled, however, that there was ultimately no error made as defined by the IPA.

"This is because the activity was lawfully authorised under an interception warrant and the adoption of an unintentionally and unforeseeably imperfect target discovery methodology did not constitute such an error," the report stated.

Despite the agency's vast technical expertise, IPCO said that the issue affecting the software was not foreseeable.

GCHQ was not the only offender among the UK intelligence community (UKIC), with MI5's mishandling of the infamous Agent X case taking center stage in the report. Agent X was an anonymous MI5 informant whose real identity has never been publicly released.

A secret inquiry by IPCO as well as the report revealed that MI5 managed Agent X despite knowing he was a violent misogynist obsessed with extreme violence.

Lord Leveson said "the case of Agent X is as serious as anything I have dealt with as Commissioner," after MI5 gave false evidence to the High Court and Investigatory Powers Tribunal to protect a foreign neo-Nazi informant who was under police investigation for abusing his partner, referred to publicly as "Beth." ®

Accenture is using AI across its recruiting process, although it maintains that final hiring decisions remain in human hands.

Speaking to at its recent investor day, the global tech consultancy and outsourcing biz said it was using AI agents to sift through resumes (or CVs if you're in the UK) and schedule interviews.

Kate Clifford, global chief leadership and HR, said it was helping clients adopt a similar application of the nascent technology.

"We call it frictionless recruiting, which is really our agentic recruiting solution. We put in an agentic recruiting solution where agents now do all the interview scheduling. Agents can screen a resume. We have assessments that deeply understand the skills of the candidates to help that process move much, much faster. Now a human always makes a hiring decision," she said.

Clifford said the agent technology was used to help recruitment staff get through their work more quickly. "Our recruiters needed to spend time with candidates. They needed to spend time sourcing, but where they were actually spending their time was scheduling interviews, where they were actually spending their time was typing information into a system," she said.

The executive claimed the agentic process reduced the time taken to hire for roles by 28 percent and improved candidate quality by 30 percent.

AI agents also help new recruits get up to speed. Whereas it once took six months for an entry-level hire to become productive, the time has been cut to six weeks using AI agents, Clifford claimed, allowing them to produce billable work more rapidly.

Accenture claims to serve around 9,000 clients — including some of the world's largest businesses and government departments — with tech advice and services. For example, it won a contract with the US Internal Revenue Services worth $2.6 billion in 2023. Any tech pro looking for a job with the company is likely being screened by AI as part of the process.

The company also expects internal staff to get on the AI bandwagon. A memo sent to senior staff in February informed them that promotions to top roles at the corporation would necessitate "regular adoption" of AI tooling, and it is tracking usage.

But not everyone is convinced by the value of AI in recruitment. A study published in 2022 found the idea that AI-powered recruitment software creates a fairer process was unfounded. A paper published in the journal Philosophy and Technology demonstrated that the software is little more than "automated pseudoscience."

Tech job-hunters applying for a lucrative role at Accenture will be hoping the technology has moved on since then. ®

Two insurers are teaming up to launch what every high-tech homeowner may soon need: household insurance for robotic mishaps.

Humanoid robots are tipped by some to be the next big technology trend, with industry figures like Nvidia's Jensen Huang predicting they'll be everywhere from factories to the home.

But as with anything AI-driven, things can go horribly wrong, especially when automatons are roaming a home carrying out tasks. They have even been labelled as potential "botnets in physical form."

The timing may be apt, then, for the UK's first domestic humanoid robot insurance policy. Specialist insurer Adrian Flux is offering to cover British customers as part of its home insurance range, underwritten by Texas-based Tokio Marine HCC.

The two firms say the product is meant to get ahead of an emerging market, with domestic robots expected to appear in UK homes from 2027. Demand has already appeared in the US, where robotics and AI companies have launched humanoid robots for the home, with early models costing around $20,000.

These are less like the near-human elderly care robot in the BBC comedy series Ann Droid, and more like vaguely human-shaped machines.

The policy covers damage to, and theft of, the robot itself, plus any damage it causes to the home or belongings. It also includes personal liability if the android injures someone or damages their property, and emergency recovery or safe disablement of the humanoid after a sudden or unforeseen malfunction, such as - perhaps - attempting to chop off your fingers or put your cat on the stove.

"As robots start to arrive in people's homes, we wanted to provide clarity on who is responsible if something goes wrong," says Gerry Bucke, general manager for the insurance biz Adrian Flux. "Like the driverless car policy we launched in 2016, it will be updated as the technology and the risks evolve."

Beyond the home, investment is pouring into humanoids across a range of applications. Chinese carmaker Chery using them as showroom sales staff, while another firm has demonstrated them in automotive manufacturing.

Elon Musk famously predicted thousands of Optimus robots in Tesla factories by the end of 2025, but still believes its humanoid robot will be the company's biggest product. ®

A UK-based reseller of Shelly smart home equipment has warned customers that their payment data may be compromised after "unauthorized software" was found on its website.

"After receiving a warning from our payment provider, Stripe, we investigated and found unauthorized software on our website and server," the email from Shelly Store UK, seen by The Register, stated.

"We removed the identified malicious software, strengthened access security, applied security updates and introduced ongoing monitoring. We have also reported the incident to the Information Commissioner's Office."

Shelly Store UK informed customers it was unable to determine whether their data was accessed or copied, although all customers who made orders during an unspecified window are being contacted.

One punter told us that hours before they received the email from Shelly Store UK, they began a payment dispute with their card provider after a £221.28 transaction, dated October 26, was made in Australia using their details.

They suspect their full card details were swiped at some point between October 12, when they made their first and only order with the company, and October 27, when they received the breach notification.

Shelly Store UK did not immediately respond to our request for more details.

Its notification email stated the internal investigation remains ongoing, and in addition to payment card details, customers may have had their names, billing addresses, and contact details compromised.

Shelly Store UK recommended that customers contact their card issuer for guidance on whether they will need a replacement card, and to scour their transaction histories and report any suspicious entries.

It also urged caution in receiving communications asking for data such as payment details and security codes, and recommended a password reset.

"We're sorry for the concern and inconvenience this may cause," the company said. "We are taking this seriously and will contact you if our investigation identifies further information relevant to you."

Shelly Store UK is an authorized reseller of Shelly smart home products and is wholly independent of the main Shelly company.

"Shelly Store is a reseller of Shelly and is not part of, owned by, or directly affiliated with Shelly Group," a Shelly spokesperson said. "We do not manage the Shelly Store website, its business operations, and we do not direct our customers from any of our websites to Shelly Store UK." ®

Google is paying a higher price for its bet on AI than even Alphabet's $51.4 billion in Q3 capital expenditures suggest.

The biz appears to have traded its high-performing search advertising business for the mediocre monetization of AI Overviews and AI Mode, a misstep that contributed to investor concerns as Alphabet lost roughly $500 billion in market value despite management's attempts to reassure Wall Street.

Google's year-over-year search revenue growth slowed from 17 percent in Q2 to 12 percent in Q3. It's still headed in the right direction, but at a slower rate.

During Alphabet's conference call for investors, Douglas Anmuth from JPMorgan Chase & Co asked whether that's because Google is showing fewer ads in AI Overviews and AI Mode than it shows in its traditional search results.

Chief business officer Philipp Schindler cited several reasons for that: foreign exchange rates, Europe's elimination of its customs-duty exemption, and "the lower monetization we were seeing in AI Mode."

To compensate, Google said it is expanding the distribution of AI Overview ads and adding multiturn ads in the US for AI Mode queries.

But some Google Ads customers appear to be more concerned about quality than quantity. In a social media post, Max Anderson, CEO of OnlyFinder, complained that Google has responded to declining search volumes by attempting to manufacture revenue growth through "short-sighted, highly extractive, customer-hostile tactics," such as charging advertisers more for lower-quality clicks, including clicks on search terms they had not agreed to target.

Eric Seufert, an analyst with Mobile Dev Memo, dismissed the doomsayers in a blog post.

"It's tempting to look at the successive deceleration in Search revenue growth from Q1 2026 as a response to pressure from other chatbots and to assume that Search query volume is migrating elsewhere," he wrote. "But the earnings release notes that the 'year-over-year Search query growth rate [was] the highest it has been in the last five years' in Q3, so it's clear that Search usage isn't declining."

So people are still searching, and the business is still growing, but less convincingly, thanks to the lower monetization of AI Overviews and AI Mode. And Google's answer involves more ads in AI interactions.

Nate Elliott, principal analyst for AI at Emarketer, told The Register in an email that AI represents a double-edged sword for Google's ad business.

"On the one hand, Google owns the thing that people use AI for most: search," Elliott wrote. "Our surveys say that by far the top reason people use AI is to look for facts and information. Luckily for Google, people also tell us they haven't stopped using search engines - they've simply added AI as a new behavior on top of search. That means people spend more time than ever looking for information, and most of those people don't even think twice before doing their searches on Google. If Google figures out how to effectively monetize those users, that's great news for them."

But, said Elliott, people may find tools they like better than Google Search as they become accustomed to AI-based discovery.

"If ChatGPT or Claude or Muse is better at helping them find what they're looking for, people's search habits might change for the first time in a generation," he said. "That's potentially catastrophic for Google's ad business. The same is true if they maintain their market share but simply can't figure out how to make money from this new kind of discovery.

"The problem is, Google isn't giving us very much information about what the specific impact is of AI on their business right now, so everyone's reading into the tiny scraps of information they are releasing."

Elliott said it's remarkable that Google has managed to keep the largest and most mature digital ad business growing at a double-digit rate every quarter in recent years.

"But it appears that most folks disagree, and instead are concerned that the advertising business 'only' grew by 10 percent," he said. ®

Welcome to another instalment of On Call, the reader-contributed column The Register publishes each Friday as a cathartic service for readers who know that dealing with tech support BS can be a cow of a job.

This week, meet a reader we'll Regomize as "Jason" who told us he had helped an agricultural equipment dealer with his computer issues for about 20 years.

"One day this guy called me to ask if I could help at a farm in Yorkshire, where he'd arranged the supply and installation of a fairly substantial system that used a computer to do effectively everything for a few hundred cattle," he told On Call.

"That meant individually feeding them, measuring milk yields, planning pregnancies, and periodically submitting all the data to the government."

Jason said other tech support folk had tried, and failed, to fix the system.

"This had somewhat soured the relationship between the agricultural equipment dealer and the farmer," he punned in his email to On Call.

Well punned, Jason.

But we digress.

"Despite not being fond of wading around in cow shit at four in the morning, I agreed to take a look at the system," Jason told us. By this time he had learned that the farm produced many tons of milk each week, so assumed he would find serious industrial-scale tech once he reached the farm.

That assumption was wrong.

"When I arrived, the first thing that I found to worry about was that the whole thing was controlled by proprietary software on a single PC running Windows XP - just after Microsoft ended mainstream support."

The other things to worry about were all acts of omission: the farm had no backups, no spare PC, and nobody on site to maintain the PC.

Jason asked if anyone was responsible for keeping the computer up to date with the security patches and was told the farmer's 14-year-old grandson had done some computer studies at school, so he got the job.

That lad had either not discovered, or deleted, the many rather racy videos Jason found on the PC.

"When the farmer's wife learned of this, there was what can best be described as a hullabaloo," Jason told On Call.

As he dug into the mess, Jason found that the computer was sending milk production data to Her Majesty's Government. He also found that neither the government, nor the farm management software vendor, had bothered to require installation of a firewall.

Jason says he installed one, made a few other tweaks, and got the system working.

"Everyone seemed pleased, or at least they did until they got my bill," he wrote. "And to be fair, they did pay it promptly."

Jason never heard from the farm again.

"I don't know if that's because it's still working OK, or if they're just too embarrassed to see me again, or they found someone cheaper who could cope with my recommendations," he wrote.

What's the strangest or dirtiest place you've been asked to deliver tech support? Don't cheese us off by holding back! Click here to send us your story so we can milk it for all it's worth in a future instalment of On Call. ®

Amazon Web Services will become a trillion-dollar business in five to ten years, and thanks to the AI boom Amazon CEO Andy Jassy thinks it should be able to grow by 40 percent or more in 2027 or 2028.

The cloud giant is already on track to generate $184 billion annual revenue this financial year, helped by 39.3 percent year-over-year growth that propelled Q3 revenue to a whisker short of $46 billion.

"It's important to remember that the vast majority of global IT spend, about 85 percent, is still on premises, and we're in a 10- to 20-year cycle where that equation will flip to the cloud," he said on Amazon's Q3 earnings call.

"Conventional wisdom has been that linchpin workloads like mainframes or storage area networks would never move to the cloud," he observed, before triumphantly noting "these workloads are moving."

He thinks AI will help, too.

Jassy said the cloudy concern installs a new AI accelerator "about every 14 seconds, and almost as fast as we install capacity, customers are gobbling it up." The outfit's Bedrock AI services is already the fastest growing multi-billion dollar service in its history, Jassy said.

Core compute and storage services are also soaring: the CEO said revenue from both "grew at their fastest rates in several years."

Jassy thinks the performance of Bedrock and core services are intertwined, because when users select AWS for AI workloads they want their cloudy inferencing rigs to be "near the rest of their data and apps."

The CEO said AI workloads therefore "can't help but drive core AWS demand because AI uses core for work like reinforcement learning, post-training, and tool usage, which can't help but drive reinvestment back into our broader AWS functionality."

He thinks this scheme is already working because margins for AI services are already "well ahead" of the figures AWS was able to achieve for its core services at a similar time in their evolution.

The company expects to make $220 billion of capital expenditure in 2026 and already spent $75.9 billion last quarter alone, most of that on AWS. Jassy said the cloud colossus is finding ways to more quickly pay for its infrastructure investments.

The CEO said it's not easy to do that.

"I think what we're seeing with AI, it's the biggest transformation in our lifetime," he said. "Would it be more convenient if this type of acceleration and opportunity happened over 25 years and you could spread the CapEx over 25 years? Yeah, it might be more convenient, but that's not the world that we live in, and that's not the opportunity."

"We can fund this plan. I think it will forever change the size of Amazon. It will forever change our revenue, it will forever change our free cash flow, and we're going to invest, and we're going to pursue it to continue to be the significant leader like we are today."

Amazon needs AWS to land this plane, because the cloudy concern is growing much faster than the company's retail and operations, which saw revenue rise six percent on far slimmer margins. Jassy could at least point to Prime Video becoming "a significant business with a revenue run rate in the tens of billions of dollars, [and is] now profitable in its own right."

Overall revenue was $180.17 bilion for the quarter.

So if the respective growth rates of AWS and the rest of Amazon continue, in a few years we could be talking about a cloud company with a retail side hustle. ®

Crypto exchange Binance has paused trading after discovering unknown parties cracked one of its workers' laptops and used it to make massive trades

Late on Thursday, US time, the company used its social media feeds to report "Our teams are aware of reports of a potential security issue and are actively investigating."

The exchange insisted "User funds are completely safe" and warned its customers "should remain cautious of scams and should not trust unofficial links, messages, or requests for account information."

The company soon froze $2.5 billion worth of assets and warned that the figure could go higher "as our teams work closely with law enforcement, issuers, blockchain intelligence firms and industry partners to trace and freeze additional assets."

To prevent the matter worsening, Binance halted all trading, deposits, and withdrawals. It also promised to cover any losses.

About an hour ago at the time of writing, Binance posted an update on the incident that explains what it thinks went wrong.

"Binance's initial investigation suggests that the security incident may involve a compromised DevOps engineer's laptop, which allowed an attacker to gain access to a service container and submit unauthorized withdrawal transactions," the post states.

The company also hiked the amount it thinks is at risk to $6.3 billion and repeated its promise to cover any customer losses.

The update also says Binance found no evidence of any wallet private key leak, suggesting that customers' holdings remain secure.

When they can use them again, as Binance is yet to resume trading.

Binance hasn't explained how attackers gained access to the engineer's laptop. The scant details provided make it hard to rule out a successful phishing attack, or perhaps a supply chain attack on a package the developer used.

Binance says it will spill once it knows exactly what happened.

Cryptocurrency outfits have a long and inglorious history of falling victim to cyberattacks, perhaps starting with the 2011 raid on Mt Gox, to incidents allegedly perpetrated by nation-states, and, just last week, another situation possibly involving an AI-powered attacker. ®

Intel's share price sank by almost a quarter on Thursday after the company revealed a deep dip in sales of PC processors.

The rout began after the silicon-slinger posted Q3 results that included some nice numbers, such as a 21 percent year on year revenue jump to $16.5 billion, helped by 78 percent growth for datacenter and AI products, and gross margins of 43 percent. The revenue number was at the high end of Intel's revenue guidance of $15.8-16.8 billion, and the gross margin beat its predictions by two points.

Outweighing those upsides was a decline in revenue for client computing and physical AI products, which saw incoming cash reach $8.1 billion but fall five percent year-over-year and nine percent decline from last quarter.

In prepared remarks [PDF] published to coincide with Intel's earnings announcement, CEO Lip Bu Tan said "In recent weeks, our customers began signaling a PC slowdown based on memory prices and availability. Year-to-date system end-point prices have increased significantly placing additional pressure on already limited IT budgets."

"The pace of the PC correction has accelerated over the last two months, as our OEM partners are aggressively managing inventory to match the softer sell through," he added, before predicting that the market for PC chips will shrink by 12 percent next year and remain tough "until memory availability and prices begin to normalize."

Which may not be for years, given Korean memory-maker SK Hynix this week predicted that memory shortages will persist until the year 2030. Intel's biggest business is therefore in trouble.

Tan tried to give investors reasons to be cheerful, by saying Intel sees a "resurgence in demand for CPU" sparked by customers' desire for computing systems that consume less energy.

"Increasingly, energy efficiency is becoming the most important metric for the AI infrastructure build-out," he said.

That matters, Tan argued, because "While GPUs are great at 'thinking tasks,' deterministic CPUs are far more efficient at 'finding' tasks. While a GPU reasoning query can run as high as multiple watt-hours of energy, a single CPU-based query could cost only joules of energy."

The CEO reasoned that lower energy consumption is important in the context of agentic AI - the idea that we will all soon run many autonomous agents to handle personal and professional scutwork.

"Applying even conservative estimates to the total number of digital agents, and agents serviced per CPU, it is easy to anticipate a massive dollar and unit TAM [total addressable market] for CPUs over the coming five years," Tan said. "By our estimates, server CPU TAM, largely driven by agentic workloads, could increase as much as 5X from where it is today, from $40B to more than $200B over five years."

But during its earnings call, Intel execs admitted it may struggle to cash in on that boom in the short and longer term as the company struggles to source key supplies like silicon wafers.

Asked if Intel can meet demand for server CPUs and if investors should therefore expect further growth in datacenter product sales early in Intel's next financial year, chief financial officer Dave Zinsner declined to offer any guidance.

"There is going to be a strong driver of demand for CPUs," he demurred. "We see that continuing all through next year. We are working on the supply side of the equation to make sure that we can meet what customers need. We will likely not fully meet what they require, but I think we are going to be in a more challenging situation in the first half of the year than we are in the second half."

Intel is trying to shift production around its fleet of fabs so it can crank out as much silicon as possible and will even devote some PC chip production capacity to datacenter products. But overall, Chipzilla is clearly in a spot of bother.

CEO Lip Bu Tan said his plan to "rebuild" Intel will be "a multiyear journey" and insisted "We have made enormous strides in these 18 months, including a dramatic re-sizing of our workforce, eliminating middle management and bureaucracy, streamlining our executive ranks and establishing an almost entirely new leadership team."

He pointed to strong progress on Intel's manufacturing processes that will bring it closer to parity with TSMC, meaning its own products will become more potent while the company's foundry business also becomes more attractive.

And he said that Intel has restored an "engineering-centric culture" that will help the company to improve.

But the CEO also admitted "I am disappointed that we were not able to meet strong customer demand in this quarter."

"We face near term challenges, including manufacturing supply shortages, the unprecedented memory crunch, and the resulting softness in PC demand. But I remain confident in our path forward."

Investors did not share that confidence. Intel's shares started the day at $92.96 apiece. Immediately after the company announced its results, shares sank to $73. At the time of writing, after-hours trading has seen the price reach $70.87, a 23.7 percent slump. ®

Multiple vulnerabilities in five popular AI developer tools could allow attackers to execute code on a host machine, read and steal files, or spoof browser address bars. Researchers say flaws in OpenAI's Codex and Microsoft's VS Code remain unpatched, potentially putting developers who use affected configurations at risk.

AI security startup Forever Security discovered vulnerabilities in Cursor, GitHub Copilot, Anthropic's Claude Desktop, VS Code, and Codex, and disclosed four of the flaws to the affected vendors. GitHub independently fixed the Copilot vulnerability without receiving a report from Forever Security. All five share a common security weakness: embedded browsers granted capabilities beyond those available to conventional browsers.

Each tool "enriched its embedded browser beyond what a normal browser allows, and those enrichments undermined the traditional threat model - as agent input, as the authority on a tab's own address, or as reach onto the host," Forever researcher Gal Weizman wrote in a Thursday report.

Forever dubbed this class of vulnerability "Embrace," short for EMbedded BRowser, and said the flaws led to code execution in Cursor, data theft in Claude Desktop, Codex, and VS Code, and address bar spoofing in GitHub Copilot.

Here's what each of these looks like, starting with Cursor.

Typical web pages can't run commands on the host computer because a web page's JavaScript cannot talk to local apps.

Cursor's integrated browser, however, allows a page to call the agent's composer. The agent reads this as instructions and follows them, but the instructions remain hidden from human users. This opens the coding agent up to prompt injection attacks that could allow an attacker to execute commands on the host machine.

The researchers disclosed the vulnerability to Cursor on September 10, and the vendor silently patched the flaw on October 1. According to Forever, Cursor did not publish a security advisory about the issue or respond to the researchers' disclosure.

Cursor also did not reply to The Register's request for comment.

Under normal browser operations, a webpage cannot change the content of the address bar.

"Copilot's browser preview works differently," Weizman wrote. "Some pages change their address without loading a new page, and the app cannot always see that. So Copilot let the page tell the app its own address."

The app checked only whether the text looked like a web address, starting with http, https or file, but it didn't check where the tab really pointed to. This can be abused to trick the embedded browser into displaying a spoofed URL - such as github.com/login - in the address bar, while loading a fraudulent website - like a phishing page - underneath.

The researchers found the address-bar spoofing flaw in GitHub Copilot 1.1.27 on September 26. GitHub independently fixed the flaw in a later Copilot build (1.1.29) on October 13 without receiving a vulnerability report from Forever.

We reached out to GitHub and did not receive a response.

All three of these vendors' coding assistants allow local HTML files opened in the built-in browser to escape its boundaries and access other files.

Both Claude Desktop and VS Code allow local pages access to their respective project folders, but the researchers found that these restrictions did not apply to web workers - a browser feature that allows JavaScript to run in a background thread, separate from the page's main thread.

This means a local page can start a worker, and the worker can bypass restrictions to reach any file on the machine.

Codex, however, didn't have this same restriction, meaning a local page could read any file directly without a worker.

Weizman said his team disclosed the issue to Anthropic on July 23, and the AI vendor declined the Claude Desktop bug report as invalid for a "workspace trust" gated security boundary. A day later, however, Noma Security reported the same flaw, according to Weizman, and Anthropic accepted the report as valid.

Anthropic did not respond to The Register's questions about either bug report or Forever's timeline of events.

Forever disclosed the Codex bug to OpenAI on July 27 and didn't get a response. The security team says it's still unpatched. OpenAI did not respond to our inquiries, including if it planned to fix the flaw.

On September 10, Forever disclosed the VS Code vulnerability to Microsoft and says the vendor also labeled it as an invalid workspace trust gated security boundary issue. The researchers claim it remains unpatched. Microsoft did not respond to our request for comment.®

If you want to ask AI a question and watch the conversation go up in virtual flames when you're done, there's a new Mac app for that.

Howler is an open-source app from a Chinese developer going by Aries Zhou, and it takes incognito mode AI questions to a new level by making them graphically interesting and easy to access right from the desktop of a macOS device.

Anyone familiar with the Harry Potter series is likely to know where the name comes from - Howlers are letters that scream in recipients' faces and then burst into flames, destroying themselves, which is the exact same thing the prompts ostensibly do once a user taps the Escape key.

The Howler app displays a parchment-esque letter on the screen whenever the user presses Option+H, complete with a little wax seal that shows a letter representing their chosen model. Responses to queries are written out "at a hand's pace," per Howler's website, first in glowing gold that settles to a sepia-toned print.

"Besides needing translation at any time, I often want to look up a concept or learn about an event on the fly, without wanting to switch to a browser or open a separate session in Chat," Zhou wrote in a comment on his Hacker News post showing off the applet. "So I made Howler, dedicated to Harry Potter fans, always there when you call, burning after reading."

The biggest catch to using Howler is that setting it up involves either using an AI service that can run commands, like Claude Code or another coding agent, or running the configuration commands manually. The recommended method involves pasting a setup prompt into the agent's input field. Once set up, Howler can connect to supported AI models using an API key, or to compatible models running locally. Howler's website states that the API key never enters the chat and is stored in the macOS Keychain for safety.

In addition to the API key for the chosen model staying local and not being printed, Howler is described on its website as only storing messages in memory with no history, sending requests straight to the chosen AI service and never through a relay, keeping no logs, and only connecting to an outside service to check for updates - and even that feature can be disabled.

It's a handy tool for those who want to ask AI questions without leaving a chat history on their Mac - like trying to remember the name of that damn function you're supposed to have down by now, or whether Memorial Day or Labor Day is the one in the fall. Just don't assume the AI provider won't retain a record of the exchange.

"Unlike regular conversations, it doesn't take up your conversation list, and you don't have to think about cleaning it up or archiving it afterward," Zhou explained. "It's only for handling your sudden flashes of inspiration or a moment of confusion."

And it does it with panache - sure, it's just an interface for an AI model one could find elsewhere, but it's easily conjured, dismissed with a snap, and is open source, to boot. You can find Howler on GitHub, or download it from its website, both linked above. ®

Vera Platforms, a startup founded by researchers at the University of Michigan, believes there's a way to make agents trustworthy enough to handle DevOps.

DevOps, which focuses on how software gets deployed and operated, is a particularly unforgiving affair. While AI agents have some leeway to make mistakes and correct errors as they generate code and work through a prompt, it's an entirely different matter when they make changes to real infrastructure running production workloads.

"Live infrastructure is unforgiving," Ang Chen, CEO of Vera Platforms and an associate professor at the University of Michigan, told The Register. "An incorrect change can take down a service, open a security hole or erase data. Mistakes are often irreversible, and there's no safe way to experiment: running code against production infra just to see what happens isn't an option."

For the past few years, Chen and his colleagues have been working on a way to benchmark agentic DevOps and a gym - a simulation environment - to help AI models get better at DevOps tasks.

"From the research that we did, we've figured out various weaknesses of leading models," said Chen. "For instance, [Anthropic's] Claude, although it's very strong with Python generation, it just has some fundamental limitations that it can't get right when it comes to DevOps."

An example of that, Chen said, is that when they asked Claude or OpenAI's Codex to decommission resources in the AWS cloud, the models didn't look beyond the default region.

"The default region is us-east-1," he explained. "There are various regions in AWS, but Claude doesn't have this awareness that the cloud resources could exist in multiple regions, so it only decommissioned part of the resource."

The fundamental reason, he said, is that these models have seen a lot of public code, but a lot less operational tracing.

Vera Platforms aims to remedy the situation with three products: vera-gym, a playground environment that emulates production infrastructure; vera-bench, a test for measuring where AI agents fumble DevOps tasks; and vera-agents, AI agents that have been trained in vera-gym and are ready to be deployed for DevOps tasks.

The CloudOpsBench test being released builds upon work done in 2024 that was presented at NeurIPS. According to Chen, Claude and Codex only score about a 50 percent pass rate. "It's a very, very hard set of tasks," he said.

The test covers tasks like deleting decommissioned S3 buckets, repairing a compliance event pipeline, and restoring DynamoDB write autoscaling. "The tasks are authored by senior infra operators with many years of experience, with rigorous peer review," said Chen.

This benchmark has the advantage of not requiring an actual cloud deployment.

"We have a high fidelity simulator that mimics Amazon behavior," said Chen. "So the rollout, the performance of the agent on the real cloud and the performance of the agent in our emulator is exactly the same. So for every API, we have this comparison and the agent sees the same thing."

Vera-agents, Chen said, should be available in Q1 2027.

In their academic work, the researchers describe cloud infrastructure as "critical but tedious work." With a bit more training, maybe AI agents can take over and keep things boring. ®

Popular for other uses, Rust isn't typically deployed for web development. That could change, however.

With AI here to do the heavy lifting, coders may want to consider the programming language, given its runtime performance and safety guarantees, suggested Carl Lerche, an Amazon Web Services principal engineer, speaking at ScyllaDB's 2026 P99 virtual conference last week.

Lerche was there to demonstrate his latest project, Topcoat, a framework for running web applications written in Rust.

Lerche is perhaps best known for creating Tokio, which is the most popular asynchronous Rust runtime in that ecosystem.

"Three years ago, had [I] gone out and said that Rust could be the best language for building web apps, you probably would have called me crazy," he told the virtual audience.

Web developers gravitate towards the languages they find most familiar. These days, that means JavaScript, or possibly Ruby.

A number of web frameworks already use Rust. Both Leptos and Dioxus compile Rust into WebAssembly, which most browsers run. This approach is great for heavy computational jobs, though the upfront loading time can be a real drag.

Topcoat compiles your Rust code into more web-friendly JavaScript. For Lerche, JavaScript remains the best way to interact with the browser's Document Object Model (DOM).

"Generally, the approach taken by any Rust framework can be traced back to another ecosystem, usually a very popular one. You can't really point to a single existing framework as [Topcoat's] main source of inspiration," noted frontend developer Ken Kohlman, in a summary video on his "Code to the Moon" YouTube channel.

Lerche admitted that Topcoat isn't for building complex applications, on the scale of, say, Figma, but it can easily be used for the run-of-the-mill, browser-based, database-driven applications, which is to say 90 percent of the web today.

Unlike most present-day JavaScript frameworks, Topcoat eschews the current definition of being reactive, where the program logic is dumped onto the browser when the session begins. Instead, most of the logic executes on the server, with only the results delivered to the browser (Developers can still write client-side logic using a subset of Rust, which Topcoat automatically compiles into JavaScript for the browser).

It's an old idea recast with better tech. Server-side rendering predates reactive programming (see Apache's server-side includes). With server-side rendering, Lerche admitted, "It might sound like we're going back to the old days of full page refreshes for every click [but] you can actually get almost all of what you think of as reactive with server-side rendering."

But server-side programming has been making a comeback lately, especially for those devs disgruntled with React and other frameworks. HTMX, Datastar, and Phoenix LiveView all render web pages on the server, sending them back to the browser nearly completed.

Topcoat is just tweaking their ideas to fit well with Rust, Lerche said. When a user smashes a button on a web page, the whole page isn't refreshed. Just the HTML snippet for that component is returned and refreshed by the server.

Topcoat saves load times by rendering all the components at once, rather than one by one. Also, data loading does not block the loading of the page. The server can send back a web page with placeholders into which the data will stream.

One big dollop of secret sauce that Topcoat has is that it provides a way for the component, which is already on the server, to speak directly to the database, thanks to an earlier project from Lerche called Toasty, a Rust-based asynchronous database object-relational mapper (which he spoke about at P99 2024).

"You can do your rendering right with the data immediately, so you don't have to figure out how to serialize it, send it to the client, deserialize it," he said.

By pushing data loading deep down into the page render, the server can send the head of the HTML template to the client immediately, allowing the browser to start preloading assets (like CSS or images) while the server renders the rest of the page.

"The simplicity is incredible, and it is something that we have been sorely missing in this space. I feel like someone who doesn't really know Rust could get off the ground pretty quickly with this thing," Kohlman said. He did agree with Lerche that Topcoat is best for smaller web projects, given that Topcoat only uses a subset of the Rust syntax.

"There's something to be said for having the full Rust syntax at your disposal," he said. ®

Suspected Chinese attackers chained PHP2Shell with a Linux kernel privilege-escalation exploit to root 25 servers, including Palo Alto Networks firewalls linked to US government-affiliated organizations, according to Dream researchers.

The snoops began exploiting these servers on October 23, targeting organizations across the US, Japan, Taiwan, and South Korea. The attackers also honed in on US government-affiliated firewalls.

"Within hours of compromising a US government-affiliated firewall, they had tunneled into the internal network, compromised a third server using the same PHP2Shell exploit, and launched a full Active Directory attack campaign - all through a chain that started with a single unauthenticated HTTPS request to a management interface that should never have been exposed to the internet," the Israeli AI cybersecurity startup said in a report.

Plus, the attackers "likely still" have persistent, remote access to compromised organizations' networks, Dream noted.

The attackers rooted the first US government-affiliated firewall on October 23, but the device crashed shortly afterward, possibly due to around 900 MB of exploit traffic. According to Dream, it remained unresponsive despite 240 automated checks over two hours. They successfully exploited the second firewall a day later, on October 24.

The compromised firewalls ran vulnerable versions of Palo Alto Networks' PAN-OS. Dream said it had been working with the vendor to investigate the compromises, and Palo Alto Networks' Unit 42 researchers independently verified the attacker's tooling and confirmed Dream's reported results.

To break into the firewalls, the attackers built a custom exploit chain that includes PHP2Shell and a Fragnesia kernel flaw, and ran about 40 custom post-exploitation scripts once they were inside the targeted organizations.

PHP2Shell, comprised of CVE-2026-109618 and CVE-2026-109619, came to light early last week, along with fears that attackers would abuse the bugs to take down most of the internet. Despite "hundreds" of PHP2Shell proof-of-concept repositories on GitHub, the first reports of successful attacks didn't start until Saturday.

Both bugs have CVSS scores of 9.2 and can be exploited without authentication to run code on vulnerable PHP servers.

The PHP2Shell exploit recovered by Dream was not a copy of any known public exploit. It "appears to have been independently developed and tailored specifically for the targeted firewall platform," the researchers wrote.

The attackers combined PHP2Shell with Fragnesia (CVE-2026-46300), a high-severity local privilege-escalation vulnerability affecting the Linux kernel's XFRM ESP-in-TCP subsystem. The US Cybersecurity and Infrastructure Security Agency (CISA) added this Linux kernel bug to its catalog of Known Exploited Vulnerabilities on Wednesday.

Both CISA and the FBI declined to comment.

The miscreants developed a four-stage PHP2Shell exploit that allowed them to execute commands without authentication on vulnerable Palo Alto Networks firewalls. They then used Fragnesia to escalate their privileges and gain root access. The attackers targeted PAN-OS 12.1 and 12.2, which run PHP 8.0.30 with a high memory limit that enabled the exploit's memory-corruption stages.

"The total chain - from unauthenticated HTTPS connection to persistent root webshell - completes in under 15 minutes," the Dream team wrote.

Here's how it works: First, they flood the servers with crafted multipart POST requests to corrupt PHP's memory, and force a predictable memory layout.

Then a second request forges the stale session ID length from the memory corruption part of the attack, forcing an out-of-bounds read and exposing the base address of libphp.so in the running process.

The attackers then exhaust PHP's memory again to force mod_data=NULL, creating a stale reference they can forge in the final stage to execute commands as the nobody user. They use Fragnesia to escalate to root and plant a web shell in a publicly accessible directory on the firewall.

After breaking in via the compromised firewalls, the attackers snooped around in Active Directory, stealing credentials and other sensitive information. With root access, the attackers also had the master key to decrypt the LDAP service account password, which they hardcoded into every subsequent script.

They also enumerated every user and group in the domain via LDAP queries, and sprayed passwords against every user account. Attackers also tried to create backdoor AD accounts and attempted SAML signing key extraction - but Dream doesn't know if these were successful.

"The attack results - whether the password spray found valid credentials, whether the backdoor accounts were created, whether SAML keys were extracted - stayed on the management server's local filesystem and were not among the artifacts we recovered," the researchers noted.

The investigation also uncovered evidence that the attackers used AI tools, but noted that this was not a fully autonomous attack.

The researchers found Anthropic's Claude Code installed, an AI-powered offensive security platform called aiscan configured with Chinese models DeepSeek and Kimi, and API keys for the FOFA reconnaissance service. FOFA is a Chinese internet asset search engine similar to Shodan. The attackers used it to identify about 19,000 internet-facing Palo Alto Networks firewalls worldwide.

"The persistent tunnel into the US government organization is likely still running," Dream warned. "And the attacker still has 19,000 fingerprinted targets they can revisit as they develop profiles for additional PAN-OS versions."®

It's a few years behind schedule, but development has resumed on the desktop edition of Canonical's existing immutable IoT distro, Ubuntu Core.

Engineering veep Jon Seager announced that Canonical is Accelerating Ubuntu Core Desktop on the company's Discourse forum. Even before his announcement, there was some fresh activity on the project's GitHub.

The post gives some explanation of why the effort is running so late, envisions three potential use cases, and talks a little about the work that is already going on to enable it to finally happen. There's no solid timeline given, but Seager says that "over the coming release cycle" the team will be working out what needs to happen, which means that "over the coming weeks, we'll have the infrastructure ready for… daily image builds for Ubuntu Core Desktop."

"Accelerating" sounds good, but with the best will in the world, it could hardly slow it down. In mid-2023 it looked like Desktop Ubuntu Core would arrive in 2024, and at that year's Ubuntu Summit in Rīga, Latvia, it revealed more details. At the 2024 Summit, there was a talk about a KDE version - but no release. At the 2025 Summit, the Reg FOSS desk interviewed Seager and "Where's Core Desktop?" was our first question.

The three roles his post discusses are the "browser-centric machine", a "developer workstation," and a "gaming machine." For this vulture, these goals imply two separate things. In general, this seems to indicate that it's wary of making something too specialized early on, then finding that it's not versatile enough for more challenging roles.

Secondly, we think that the company is thinking of three specific niches that rival immutable distros have occupied successfully for years. Right now, the "browser-centric" one is filled by ChromeOS and ChromeOS Flex. We use Flex - it's pretty good, even if its days are now numbered due to the new Android-based GooglebookOS.

Secondly, Fedora's Atomic editions have been around for years now - The Register first mentioned Fedora Kinoite well over five years ago. As well as Fedoras's own versions, there are also community-led distros based on the same tech, called Universal Blue. We've tried several, and while for now they offer nothing much this vulture wants, they undeniably work and they have happy users. Fedora's marketing aims at developers, and an immutable distro with containerized environments on top is harder to accidentally break with crashy work-in-progress code.

Thirdly, the games-machine niche is owned by Valve's SteamOS. Although it is native to the Steam Deck - and if you're among the lucky few, the Steam Machine and Steam Frame - you are also free to run this on your own AMD-based kit.

All three of these product families have been well-established for between five to ten years now. Canonical is lagging behind, and this is a subtle acknowledgement that it knows who the leaders are. We didn't doubt that he knows his stuff - a year ago, Seager's blog post about The Immutable Linux Paradox went into the differences in depth.

The delay in delivering an immutable desktop is not for lack of immutable-distro technology: Canonical's had that for years. It announced Ubuntu Core 16 a decade ago, and when we tried Ubuntu Core, that was the fourth release.

Seager's post identifies some of the problems that have delayed its Core Desktop. One is that the 2023 preview required its own, specially modified version of the base core22 snap. Obviously, the company would prefer that it was based on the existing Core distro.

He also points to some essential underlying tech that is not obvious. He mentions that the Snap Store already has over 11,000 packages, but notes that many still use weak "classic" confinement - such as VS Code. This is a problem for Core, which requires strict confinement. He also notes some less-obvious things that need to work first, including smartphone-style permissions prompting, inter-snap dependencies, content snaps and snap components.

Some of the plans are ambitious, but in good ways. He discusses the need for modularity in the desktop snap packages, so that Core Desktop won't be restricted to only GNOME. This means isolating the display manager (that is, the login screen) from the desktop, and the desktop from the browser. Here, he gives a nod to external consultancy enioka Haute Couture for its work on KDE snaps.

It's still much too easy to break any ordinary desktop Linux distro. In this vulture's extensive personal experience, the Btrfs filesystem is especially vulnerable. In four years of daily-driving a Btrfs-based distro in the 20-teens, we saw more crashes due to corrupted Btrfs volumes than in the previous two decades. However, all the vendors of Btrfs-based OSes are resolute in their denial about its real problems, which means they never get fixed. SteamOS and KDE Linux both use dual read-only Btrfs partitions, which update one another ChromeOS-style. This may be the only way to battle-harden this problematic filesystem. Canonical, on the other hand, offers integrated OpenZFS. We've been using ZFS since before the COVID pandemic, and it has yet to lose a single byte of data for us.

There are still blockers before Canonical can deliver a versatile, modular, desktop version of its existing immutable OS, but it's having another go. It's about time, and there is some new hope that we may start to see previews in the Ubuntu 27.04 timeframe. ®

America's national laboratories are leading a consortium to move advanced battery storage technologies from the lab into domestic manufacturing.

Dubbed the Battery Manufacturing Accelerator, or BattMan (no, really), the initiative aims to bring together a bunch of national laboratories, universities, and companies to overcome barriers in scaling up new battery technologies.

Novel battery storage solutions are needed for applications in the electric grid, datacenters, and electric vehicles, as electrification gradually extends across industry.

Batteries have a far lower energy density than fuels such as petroleum, so the hunt for better materials never stops. Researchers also want to improve on the safety of flammable lithium-ion cells, cut production costs, speed up charging, and reduce reliance on scarce raw materials.

"The advances we make today in battery manufacturing technology will determine America's energy security and economic competitiveness for decades to come," said Audrey Robertson, assistant secretary of Energy at the Department of Energy (DoE), which is overseeing the program.

Researchers continue to develop new battery materials and chemistries, yet many promising technologies face obstacles on the way from the laboratory to commercial production. According to the DoE, BattMan will tackle that problem by combining super intelligence - known outside of the Trump administration as AI - with advanced manufacturing into a coordinated, data-driven innovation ecosystem.

Pacific Northwest National Laboratory leads the consortium, along with Oak Ridge National Laboratory, Argonne National Laboratory and National Laboratory of the Rockies. Academic partners include the University of Washington, Binghamton University, University of Alabama, Arizona State University, and University of Michigan-Dearborn.

The accelerator is also building ties with battery and materials manufacturers, and hopes to sidestep critical minerals supply problems by finding battery materials among more abundant, domestically sourced elements.

ORNL is contributing its Battery Manufacturing Facility (BMF) R&D center and plans a smart test bed with advanced sensing, real-time monitoring, and data analysis to trial both wet and dry electrode manufacturing.

BattMan will also introduce autonomous pilot lines that combine sensing, process monitoring and intelligent control systems to improve product quality and reduce variability in the manufacturing process. The goal is an AI-based manufacturing platform that spots production irregularities as they occur and adjusts parameters on its own to prevent defects.

"The BattMan Accelerator is exactly the kind of coordinated national effort that turns scientific breakthroughs into economic and energy security advantages," claimed distinguished R&D staff member Chins Chinnasamy, the consortium lead for ORNL.

Organizations interested in becoming part of BattMan can visit the program website to learn more. ®

A major US federal agency has repeatedly left sensitive information in its Google Drive accessible to employees and contractors who had no legitimate business need to see it, and auditors aren't happy.

The General Services Administration (GSA), which handles contracting, procurement, property management, and other elements of basic operations for federal agencies, has faced repeated findings of inadequate protection for controlled unclassified information (CUI) since at least 2014, according to a report from the agency's Office of Inspector General. Again.

Controlled unclassified information, in the US federal government's parlance, is information that isn't classified but requires protection or restrictions on its distribution under applicable laws, regulations, or government-wide policies. Access is generally limited to individuals with a legitimate government need to know. According to the OIG report, that can include things like physical security info for government buildings, personally identifiable information (PII) of government employees, and proprietary business information, all of which the GSA has left exposed.

According to the report published this week, employees and contractors have approximately 2.5 million documents stored in GSA's Google Drive.

"During a limited search of GSA's Google Drive, we found 116 files containing CUI that were improperly shared with all GSA users," the report notes. That's a limited search, mind you, and the OIG has plenty of reason to believe there's more than just that sitting open for anyone with access to the GSA's Google Drive instance to sneak a look at.

The OIG said that GSA is failing to properly manage the third-party tool it has in place to prevent data loss and detect CUI exposures, with annual reviews of search terms incomplete or undocumented, many relevant search terms not implemented, and a failure to enable an available optical character recognition feature that could help detect CUI in scanned documents. Additionally, the GSA was apparently "unaware of the tool's file size limitation of 15 megabytes," meaning that even if the GSA was properly using that tool, some large files would still be skipped.

"We also found that GSA lacks: (1) an approval process for properly sharing sensitive information and (2) a standard naming convention that could more effectively prevent exposures of CUI in GSA's Google Drive," the OIG added.

As for what was exposed … it's not great.

According to the audit, site plans for buildings in Washington, DC were part of the improperly access-controlled files, with data that includes entrances and screening techniques, as well as the location of sensitive areas, like gun lockers. A progressive collapse analysis that "describes the building's collapse vulnerabilities, defense vulnerabilities, and critical failure locations" was also exposed.

PII of government individuals was also exposed, including social security numbers, addresses, dates of birth, and contact information. Proprietary business information was also exposed, "including proprietary research and vendor proposals with technical and pricing information."

This might not have been publicly accessible, but it's still a shocking lapse of data control, especially given that contractors with access to the GSA's Google Drive could potentially view proprietary information belonging to competing vendors, giving them an unfair advantage if they used it.

To make matters worse, this isn't the first time the GSA has been dinged by its own inspector general for such lousy data protection practices. According to the report, it was 2014 when the OIG first spotted such bad habits.

"In August 2014, we first reported that GSA did not protect sensitive information on child care centers, courthouses, water sources, building automation systems, and security and fire alarm systems in its Google cloud computing environment," the OIG noted. "We issued additional reports citing concerns regarding GSA's ability to protect sensitive information in its Google environment between 2015 and 2017."

Auditors flagged another set of improperly shared Google Drive files in April 2025. While the GSA's chief information officer agreed with the auditors' seven recommendations to fix the problems, the report doesn't establish whether the corrective work has been completed. We reached out to the GSA and the OIG's office for an update, but didn't hear back from either. ®

Oracle expects to earn more from existing application customers as it embeds AI in its enterprise software suite.

Speaking at Big Red's investor day on Wednesday, Chris Leone, executive veep of application development, said Oracle will layer usage-based revenue on top of its subscription model.

"The more work that gets done by AI in our system, the more revenue we will see because they're consuming AI units as that work is getting done. We don't have to add more seats in order for us to now generate more revenue," he said.

Earlier this month, Oracle revealed its Fusion Claw runtime, designed to help customers control AI agents that automate business processes. It targets ERP, finance, and supply chain management in cloud-based Fusion applications running on Oracle Cloud Infrastructure (OCI).

Customers will pay for more work done inside Oracle applications, and Leone said Fusion Claw will also automate work done outside of Fusion.

The promise of more revenue for Oracle, and higher bills for customers, came as the company sought to reassure investors about its spending, borrowing and revenue as it chases a share of the AI market.

Oracle is funding a datacenter spending spree with negative cashflow and borrowing. OCI is set to host large language model builders, notably OpenAI, alongside its own customers' workloads. Oracle spent $55.7 billion on datacenters and cloud infrastructure in fiscal year 2026, which ended in May, and plans to spend between $90 billion and $95 billion in fiscal 2027.

That will require more debt. In 2027, Oracle hopes to raise around $40 billion in debt and equity, on top of the $18 billion in debt it raised in September 2025.

In a notice to investors this week, Oracle said it expects fiscal 2030 revenue of $225 billion, a compound annual growth rate of 35 percent from 2026. It expects positive cash flow by the second half of fiscal 2028, with capital spending falling the following financial year. Investors reacted positively.

Oracle co-CEO Mike Sicilia also told shareholders the company will use AI to help implement its software, a plan he called as important as putting AI in the software itself. He argued AI will cut implementation time, bringing existing and new customers to OCI faster and boosting revenue.

"On-premises customers can now move far more easily to the cloud by leveraging the AI tool sets to automate migrations and decrease the cost and time to implement. They don't have to be Oracle customers; they can be competitive customers."

Sicilia said enterprise application software users' reluctance to change had been a major "headwind" for Oracle, one he hopes AI will ease.

"Our large remaining base of on-premises customers are going to start to adopt this technology quite quickly. When we move customers from on-premises to cloud, we see a multiple increase in revenue," he said. ®

Michael Stonebraker introduced Postgres® at Berkeley in 1986. Forty years later, the database he invented is a leading choice for a generation of AI projects, and he is among the sharpest skeptics of how those projects are being built. Bruce Momjian is cofounder and core team member of the PostgreSQL Global Development Group, and he works at EnterpriseDB (EDB). In upcoming episodes of AI & Data Horizons by EnterpriseDB (EDB), the Turing Award winner and MIT professor and the Postgres community advocate both make the case that the future belongs to systems that bend without breaking, and to communities that no single company controls.

Stonebraker's first exhibit is text-to-SQL, the promise that a business user can ask a question in plain English and get an answer from the warehouse. Public benchmarks such as Spider and BIRD report accuracy above 80 percent. His team tried the same thing on MIT's own data warehouse—1,400 tables of student, faculty, building, and salary information—using real queries from its SQL log.

"We got zero-percent accuracy. Not 80, not 70, not 40—zero," he said. Adding agentic AI, retrieval, and prompt engineering lifted that level into the teens. Telling the model every table and join term reached the 30s. "Still not ready for prime time."

The reasons are the stuff of every enterprise estate. There is no training data, because the warehouse is not on the public internet. The data is idiosyncratic: among other things, MIT buildings have numbers, not names. The schema has rotted into cryptic table and column names and overlapping views. And real queries are harder than benchmark ones. His team has published its own benchmark, Beaver, and says nobody has broken out of accuracy levels in the 30-percent range.

The deeper problem, Stonebraker argues, is architectural. Large language models are good with text, but "what they are no good at is doing joins of structured data," which is most of what an enterprise holds. Working with the city of Munich's transport department, Stonebraker's team found that answering a single citizen complaint about crosswalk timing meant combining regulations, CAD drawings, trolley schedules, and school-zone maps. Orchestrating that with a language model at the center was, he said, "ridiculously expensive."

The team's answer, a system called Rubicon, uses models only on text, then casts the results into tables and hands the rest to a tabular engine. The orchestrator "looks like a query optimizer." Stonebraker says it is cheaper by a factor of 10 and considerably more accurate than agent frameworks built around a model.

He expects agentic AI to slide from the top of the hype cycle into the valley of despair before a more balanced view emerges. Enterprises hate non-determinism: Approve a loan and reverse it on review, and "that's called a lawsuit." Organizations also need protection from rogue agents. The guardrails inside the database, he notes, largely exist already: decades of access control, complete logs, and triggers that can stop a transaction in its tracks. Data-heavy agents, he suggests, belong inside the data layer as user-defined functions, if the tooling to debug and version them improves.

Stonebraker is equally blunt about the frontier. The models have already used the real training data on the internet, and he knows of no enterprise that will let a model train on its proprietary data. AGI, he said, is "basically marketing hype" from foundation model vendors justifying their valuations.

Why Postgres? "Postgres is the intergalactic answer to open source," Stonebraker said. Its unique feature is that "it's not controlled by any one company." EDB, Microsoft, Google, and others each contribute committers, and nobody steers. He dates its rise to Oracle buying MySQL, when "everybody said: vendor control." For enterprises, that independence is the foundation of sovereignty: retaining control over how and where their data systems operate.

Momjian describes the same structure from the inside. Postgres has no office, no assets, and no legal entity, yet it produces "software better than billion-dollar companies." Its forecasting tool is the community itself, which spots needs early from China, Russia, and the United States. "We only need one person in the world to be supersmart, and that person to tell us what they think."

Both experts describe a design philosophy that favors flexibility over the last few percent of speed. Stonebraker built Postgres to be extended, and that is why it absorbed JSON, full-text search, GIS, and now AI without a rewrite. Momjian adds a second call from the 1990s: Lean on the operating system rather than absorb it, which made Postgres a natural fit for the cloud.

Momjian resists optimizations that buy 5 or 10 percent at the cost of rigidity, since a mistuned system can lose 40 percent. "If you take a Formula One car to the supermarket, there's no place for the groceries." The project plans on a horizon few vendors can match. "We're looking at what would be the impact 5 years from now, 10 years from now."

The new release puts that discipline on show. Several previewed features, including SQL property graph queries and partition merging and splitting, were reverted, because reliability comes first. AI-assisted review is finding problems quickly; the August 2026 patch release for PostgreSQL 18 addressed 28 CVEs.

What remains is practical: REPACK CONCURRENTLY for busy tables, smarter parallel autovacuum, and sequence synchronization for logical replication. Plan Advice, from PostgreSQL major contributor and committer Robert Haas, is the community's answer to Oracle-style optimizer hints it rejected for years. "Somebody found a way of doing it that didn't have the downsides," Momjian said.

The two diverge on how fast AI will write SQL. Momjian expects agents to create databases and generate much of the code. Stonebraker says that will not happen inside the enterprise until data is documented, silos are opened, and the politics change. They agree on the destination. AI needs a relational system to give it permanence and memory, and, in Momjian's words, that is "a game-changer."

Forty years on, the inventor and the steward are betting on the same database.

Contributed by EnterpriseDB

Belgium's finance and tax agency has pulled "several" digital services offline after a cyberattack on the organization.

FOD Financiën confirmed the attack on Wednesday, saying that MyMinfin, the agency's main online tax portal, as well as other online services, are temporarily unavailable while they start the initial investigations.

"This measure helps limit the risks and makes it possible to carry out the necessary technical interventions to ensure the security of our systems," it said in a public announcement.

FOD Financiën added that "this has consequences for tax deadlines in the near future," referring to the October 31 extended deadline to submit tax filings for specific types of income, such as sole traders and earnings from foreign companies.

The agency confirmed the deadline will be extended to November 16, by which time MyMinfin, the platform through which such filings would normally be submitted, should be back up and running.

The detail offered up in the public statement reflects the early stage of the investigation; not much is known, although there is nothing to suggest data was compromised.

"Data protection is a top priority for the FOD Financiën," it said. "At present, there are no indications that a data breach has occurred or that data has been compromised. This aspect is also being closely investigated and followed up.

"In the meantime, a large part of our other applications remains available with full or partial functionality. We are closely monitoring the situation and doing everything we can to make the affected services available again as soon as possible."

Spokesperson Francis Adyns told local news outlets that the attack was thwarted within hours, and that it was unable to gain access to applications or data.

"The attack could not be sufficiently expanded within our environment to take over applications or gain access to data," he told Nieuwsblad. "Elements that were potentially infected were isolated. As a result, MyMinfin, Tax-on-web, and Biztax went offline, but the hacking was successfully stopped.

"The various services must be restored in an isolated environment. This requires considerable time to realize, with the necessary caution to rule out new risks." ®

Everyone's favorite Unix init system is adding support for the Unix world's favorite new programming language. What's the worst that could happen?

Don't worry, this isn't here just yet. Agent P and his trusty team released systemd version 262 about a month ago. That means that systemd 263 is now under construction, and developer Christian Brauner announced on Mastodon that his new feature has been merged.

The previous effort, which was pull request #44144, has now been split into five separate parts. Part 1 contains keyring-util and voa-util, described as "helpers for enrolling certificates into kernel keyrings." Part 2, support programs written in Rust, is the banner item. Then come a few more: bind only the headers Rust code uses, wrap the libsystemd-shared helpers systemd-keyring-setup needs, and finally, add systemd-keyring-setup.

There are several important riders to the announcement, and these are spelled out in the second PR:

"Rust is only used for new leaf components. Existing C code isn't ported and nothing that other components depend on is written in Rust."

In other words, before anyone gets too excited, this does not indicate the start of an effort to rewrite systemd in Rust. All it means is that new systemd modules can be written in Rust.

The second bullet point will be more significant for existing Rust developers: "No cargo. Everything is built with meson."

Cargo is the Rust language's package manager and build system. It's the normal way to build Rust applications, as well as to distribute modules of Rust code for other Rust developers to use. However, systemd already has a modern build system. It's called Meson and it's not unique to systemd - it is also used extensively by the GNOME project and the Mesa 3D graphics library. Meson itself is written in Python and uses Ninja to manage the compilation process. Ninja was designed to build Chrome faster, and it's also used for Android.

The third stipulation is that any Rust code must not use the standard library. For those more familiar with C, this may sound extreme, but it's normal for embedded Rust programming.

The project documents the system requirements. It needs Rust version 1.85 or later. That's modest: that version was released in February 2025, and version 1.99.1 appeared earlier this month. It also calls for bindgen version 0.71.1 (from December 2024) or newer, and Meson 1.4.0 (from March 2024). These are relatively old versions - for instance, it's the Meson version included in Debian 13).

Brauner added an interesting little coda in a second Masto post: saying that the team would have liked to support Rust in the core of systemd, but currently, Rust's support for dynamic libraries isn't quite there yet.

It's perhaps worth noting that writing systemd modules in Rust isn't the same thing at all as calling systemd from Rust code. There are several tools for doing that, and they have been around for years: for example, the systemd crate first appeared in 2015. ®

United Launch Alliance (ULA) has set a target date for its return to flight: Friday, November 13, for a Vulcan launch.

The mission will send 40 Amazon Leo satellites into orbit. A pair of BE-4 engines will power the booster, along with six Solid Rocket Boosters (SRBs). A Centaur V second stage, propelled by two RL10C-1 engines, will complete the journey to orbit and deploy the payload.

The launch is the first Vulcan launch for Amazon after eight deployments with ULA's Atlas V. It is also the first since the launch earlier this year for the US Space Force in which an SRB burned through its nozzle during the ascent, treating viewers to some impressive fireworks and controllers to some likely clenched sphincters thanks to a "significant performance anomaly."

In the event, the US Space Force payload successfully reached orbit, but the Vulcan was stood down while ULA conducted an investigation. This wasn't the first time a Vulcan had had issues with its SRBs. A nozzle was shed during the second launch of the rocket in 2024. Things went OK with the third launch. But another problem on the fourth trek to orbit meant a stand-down was unavoidable.

The workhorse Atlas V has continued deploying Amazon Leo satellites during the Vulcan's hiatus. Another 29 were added on a July 2 launch from Space Launch Complex-41 at the Cape Canaveral Space Force station. However, Amazon booked 38 Vulcan rocket launches - quite an increase from the eight Atlas V launches ULA has already performed for Leo. It is obviously keen to add more satellites to its growing constellation.

The SRBs are Northrop Grumman units, and the six Graphite Epoxy Motor (GEM) 63XL units will each provide an additional 459,600 lbs (2,044 kN) of thrust at liftoff. That's assuming nothing unexpected burns through or falls off. ULA has not detailed the results of its investigation or any changes it had to make.

ULA said: "The integrated team has made great strides in improving our understanding of the solid rocket booster (SRB) anomaly that occurred on the previous Vulcan flight and have identified several potential contributors which have been mitigated through design and process improvements to the nozzle. To support our customers, this upcoming launch will utilize the improved nozzle design, and additional testing and analysis will continue in parallel with the mission. By flying the improved design, ULA gains valuable flight data that validates these improvements and is the most beneficial path in support of all our customers."

US Space Force and Amazon are ULA's biggest customers - and Space Force will be watching. ULA came second to SpaceX in contracts awarded by Space Force for FY25 to FY29. ®

The UK's comms watchdog has put forward four methods of checking netizens aren't kids - dubbed highly effective age assurance (HEAA) - as it prepares for the proposed under-16 social media ban.

The regulator says some adults may not need to prove their over-16-ness to social media platforms that will make inferences about their age based on previous activity, but confirmed that "many" are likely to have to undergo the same verification process.

Big Brother Watch's (BBW) Director Silkie Carlo said: "This plan for a government 'social media ban' is a national ID system by the back door. Both the government and Ofcom have admitted that children will get around the ban, which means the real effect will be normalizing adults and children alike to handing over their passports just to get online."

Ofcom says four methods remain from the seven it initially outlined when the Online Safety Act (OSA) came into force last year.

The seven were capable of verifying whether a subject was 18 years old, but there are fewer viable options for determining the age of 16-year-olds, it said today.

The methods put forward include:

• Photo-ID matching: Checking information displayed on identity documents and analyzing the photo against a live image of the subject taken during the verification process

• Banking checks: An estimated 90 percent of 16-17-year-olds in the UK have a bank account, which they would have provided their identity documents to open. The information processed by the bank can be used to verify whether someone is aged 16 or older

• Digital ID services: An interesting entry, given how quickly these were circumvented once the OSA age-checking requirements came into force. These third-party services analyze a person's photo once, and if it matches the age they claim to be, let them verify their age across other services

• Facial age estimation: A method that even Ofcom admitted isn't quite reliable enough to be used on its own. The regulator said that online services can use technologies to analyze a subject's facial features to determine their age. The same technologies also have a margin of error of a few years, so online services may wish to use a second method of HEAA to verify the results of this check.

The three reserved for those aged 18 and above include credit card checks (only adults can have a credit card), mobile network checks (MNOs require proof of adulthood to remove content blocks), and email checks - using a subject's email address to make educated guesses about their age based on the other types of services with which it is registered.

Ofcom acknowledged that not every 16-year-old will have a form of photo ID, which is required for all four of its HEAA methods, so they will need to acquire one in order to access online services.

It estimates that around 94 percent of 16-17-year-olds in the UK have a form of photo ID, such as a passport, driving licence, or an ID card certified by the Proof of Age Standards Scheme, and that around 150,000 have neither.

BBW's Carlo added: "Hundreds of thousands of people without ID, who tend to be poorer people, and many more who don't want to risk handing their passports over to foreign tech companies will be locked out of huge parts of the internet.

"The plan will kill anonymity online, whilst the UK's culture of speech policing is reaching a crisis point. The policy is painfully misguided, an Orwellian assault on free speech and privacy, and will make children no safer at all."

Ofcom admitted that "the risk of circumvention cannot be entirely eliminated," as some children will find ways to bypass these age checks, no matter how fully they are implemented.

It cited Australia sporadically in its announcement, a country that introduced an under-16 social media ban in December 2025, but has struggled with a large proportion of children devising ways to circumvent these checks.

Ofcom said the UK government intends to adopt, but learn lessons from, Australia's rollout, committing to implementing more robust measures that promote the ban's aims.

It said it has identified some mitigations that could be taken to counter circumvention, such as making it easier for users to report suspected underage social media users, and "using age inference models."

The regulator said it will closely monitor circumvention for the first 12 months after the ban is enforced to understand possible methods to prevent children from finding ways around the age checks.

In May, nearly a year after the OSA's age verification requirements kicked in, researchers found that the existing age checks proved largely ineffective at keeping under-18s away from adult content online.

Some reported beating age checks with just a drawn-on moustache.

England's Children's Commissioner, Dame Rachel de Souza, said recently that the OSA "has made absolutely no difference" to children's ability to access potentially harmful online content.

Commenting on today's announcement, Ofcom chief exec Dame Melanie Dawes said that when the social media ban takes effect, it will take a whole-of-society approach to ensure the ban is upheld.

?This ban will shape the future of children's lives online, so it's important to get it right," she said. "It can work if social media firms do highly effective age checks properly, and we've set out ways that could be achieved. When it comes into force next year, we'll be ready to take swift and firm action against any company that fails to comply.

"We should expect that motivated children will seek to bypass the restrictions - just like alcohol laws can't completely eradicate underage drinking - but there'll be ways for platforms to mitigate this risk. This will be a huge cultural shift for our society, so it'll take time to adjust, and everyone will have a part to play."

The UK government is expected to outline the first set of regulations related to the under-16 social media ban by the end of the year, and Ofcom is in the process of developing draft guidance for platforms. ®

EXCLUSIVE Researchers identified at least 83 attacks against 61 public GitHub repositories, including popular open source projects such as Google LangExtract, Trivy, Brave browser, and MLflow, in what they suspect is a coordinated campaign to execute commands in CI/CD environments and steal credentials and access tokens.

It all happened very quickly, the Radware threat hunters said in an exclusive interview with The Register. Most (77) of the malicious items were published during a 43-minute time span on September 1, with the remaining six occurring within the next two minutes.

"Our guess is that the final aim was a supply-chain attack in which the attacker gains credentials for the repo, and then he has the power to change the contents - the code - as he wishes" and then push the malicious changes to others, Radware VP and researcher Gabi Nakibly told The Register.

Nakibly and fellow Radware researcher Zvika Babo say they suspect a single individual is behind this supply chain campaign. This attacker used 74 different GitHub accounts, many of which had legitimate development histories, meaning they were likely compromised.

"Considering that dozens of accounts posted the same type of payload, on the same day, within a narrow time window, and to the same infrastructure, it is reasonable to assess that the activity was coordinated," the team wrote in research shared ahead of publication with The Register.

They discovered the campaign while investigating prompt injection in GitHub - and they told us the attack chain is part of what makes this supply-chain campaign novel, and extremely stealthy.

First, the attacker publishes an issue, pull request, or comment containing malicious commands hidden in HTML comments. This ensures the content is not visible in the rendered page. Plus, a GitHub event - like a pull request or issue - can automatically trigger a workflow if the repository is configured to respond to that event, meaning the attacker can be pretty hands-off at this point.

"The workflow automatically is triggered within a second, and if the workflow is indeed vulnerable, then the attack is executed successfully," Nakibly said.

The workflow incorporates attacker-controlled content directly into shell source code, and the shell executes the hidden, malicious command, which hoovers up environment variables and attempts to exfiltrate those to an attacker-controlled server.

While the researchers examined logs from actual runs, and told us they found two cases containing evidence that the malicious command ran inside the automation environment, they can't definitively say whether the attacker successfully obtained repository write permissions, modified source code, or compromised downstream software. But they're trying.

"The big picture is that the attackers are trying to find every possible way to get inside the repos to get credentials," Nakibly said.

Plus, "the attacker doesn't need to compromise the repository or to have access to its codebase" to do this, Babo added. "Basically, every user with a GitHub account can try to run this attack - it's very easy to start this attack."

As of October 13, the threat hunters identified 83 GitHub items with a verified payload across 61 different repositories. The repos span AI frameworks, security tools, development infrastructure, databases, and other widely used open source software projects.

The Radware report lists agno-agi/agno (42.7K stars), ToolJet/ToolJet (41.1K), google/langextract (38.9K), aquasecurity/trivy (38.4K), CopilotKit/CopilotKit (38K), and podman-container-tools/podman (33K), among the most popular.

Other notable targets included QwenLM/qwen-code (28.5K), mlflow/mlflow (28.4K), discordjs/discord.js (26.8K), cilium/cilium (25.6K), brave/brave-browser (23.9K), dotnet/maui (23.3K), Homebrew/homebrew-cask (22.2K), github/docs (21K), apache/doris (16K), and PX4/PX4-Autopilot (12.8K).

"These projects are used by thousands of organizations," Babo said.

The researchers disclosed their findings to all of the projects' maintainers, and most of the repositories were closed or fixed, but they can't confirm that all of them were.

This campaign follows several earlier supply chain attacks against open source security and dev tools as attackers increasingly look for new methods to steal credentials and other secrets.

"There is no way to stop the attackers from attempting," Nakibly said. "The only way is to make sure that the workflow code is resilient, and not vulnerable to such attacks."

The researchers recommend applying least-privilege access to CI/CD workflows, and separating as much as possible workflows triggered by external events from jobs with elevated permissions.

"Developers should treat external content as untrusted, and also limit the secrets and environment variables that are available to the workflow," Babo said. ®

The Document Foundation (TDF) says LibreOffice has no built-in AI features and that its core developers do not use LLMs - but community contributors face different rules.

In a Fediverse post, someone who goes by the handle of argv[-1] highlights three LibreOffice commits this year that acknowledge the use of LLM coding assistants.

Two of them - Inheritance when setting paragraph style font size to 100% or +0pt and Enable relative font sizes in Calc cell styles - date from April and credit Claude with helping identify the fixes. More recently, September's Add Copy Image button to screenshot dialog acknowledges using GitHub Copilot in VS Code to generate the BitmapTransferable class, while the copyButtonHandler method was written manually with LLM/code-completion assistance.

This is hardly a fatal flaw - after all, it's getting hard to find any major FOSS project with absolutely no LLM-generated code in it these days.

However, The Document Foundation says that the leading FOSS office suite is AI-free. As we noted when it released version 26.8 in August, we interpret this statement as a jab at its former development partner Collabora. Back in June, we noted that increased AI integration was a headline feature in Collabora CODE 26.04. CODE is Collabora's web-based office suite, based on LibreOffice code.

TDF - the primary maintainer and custodian of LibreOffice since its 2010 fork from Oracle-controlled OpenOffice.org - has made the absence of built-in AI a selling point. In September, it published a blog post declaring that no AI is now a feature. Meanwhile, on October 21, TDF announced 9.4 million downloads over eight weeks - excluding Windows updates and downloads of older releases, which are counted separately. The foundation acknowledged that the reasons for the growth are difficult to determine.

TDF's direct response on Mastodon is dismissive:

"Ah, another high-quality Mastodon discussion. Again: No core developers use LLMs. We rely on contributions from the wider community and don't explicitly list tools people can't use - but we do require people disclose when they use LLMs and generally discourage the practice. We carefully review submissions by contributors to make sure the code quality is good and that the patch author understands their own code. LLMs are prohibited for students. That is not 'slop'."

In other words, TDF's approach resembles what some commenters have compared to the Debian approach: LLM coding assistants are permitted for some contributors, subject to disclosure and human review. For AI advocates, this is the only pragmatic response - which could be summarized as the "AI is just a tool" school of thought.

However, since this admission does seem to contradict the very visible public no-AI message, some LibreOffice users are really rather unhappy. Our favorite comment so far is from Argentinian developer Federico Schonborn, who slyly noted:

"When a bakery announces they won't put asbestos in their cake frosting you must not point out they're already putting asbestos in their cake filling, you see."

We asked TDF's spokesman Italo Vignoli for comment. He pointed us to TDF's Development/AI policy, and added:

"There isn't any departure from the public statements about LO 26.8, as they were about user features and not about development."

As "Skjeggtroll" succinctly put it:

"It strikes me as odd to ban all use of LLM for the commit message, but still allow its use for the actual code." ®

The wait for a new Raspberry Pi Zero may be nearly over, judging by the raspberrypi/linux repository.

Among the merged pull requests is one titled "Add device tree for Pi Zero 3 W and Pi 4A," suggesting new hardware is on the way, including a successor to the very long-in-the-tooth Pi Zero 2 W.

The board launched five years ago. Its 512 MB of memory was hardly generous, but it offered considerable bang for the buck. At the time, we reckoned performance was closer to the Raspberry Pi 3 than the flagship Pi 4, but perfectly adequate provided you didn't open too many browser tabs at once.

According to the source tree, as reported by Phoronix, the new board could use the Broadcom BCM271, up from the RP3A0 in the current Zero. If so, that would be a significant performance gain, though it's unclear whether the 1.5GHz clock speed of the Pi 4, which uses the same SoC, will be retained.

Heat is a concern as the Pi 4 gained a reputation for running hot under load without active cooling, and it throttles performance when temperatures climb. A board as small as the Zero will have even less room to shed heat.

Assuming that challenge can be met, it would be a worthy upgrade for the venerable Pi Zero 2. There is no word on how much memory it is likely to have - we can't imagine much given the current cost of RAM and the need to keep the price at least close to that of the Zero 2 W, which can still be had for around $18.

The code also hints at eMMC storage is mentioned, but that seems unlikely to reach the finished product, probably because of the cost.

Alongside the Pi Zero 3 W is a Raspberry Pi 4 Model A. Where the Zero 3 W is a clear upgrade, the 4 Model A looks more like a cost-cutting measure amid rising hardware prices. Model A boards typically have less memory, and no Ethernet port, and the 4A appears to have a single HDMI output rather than twp.

Things could change before any hardware ships. We asked Raspberry Pi if anything was in the pipeline, but the company said it had nothing to add.

Enthusiasts will welcome an upgraded Pi Zero after more than half a decade, provided the price is right and it doesn't run too hot. At the same time, a cut-down Pi 4 carries the whiff of inevitability in an era of rising component costs. ®

Users of New Outlook wondering where the option to open an attachment went have a culprit to blame - an Edge update.

The issue underscores how interconnected things are in the Microsoft ecosystem. According to the vendor, an update to Edge 156 "unintentionally enabled an unnecessary configuration," which means some New Outlook users are unable to open email file attachments.

An error stating the file cannot be opened also appears for some, along with an instruction to restart Outlook. However, the "Open" option goes away after the error is encountered. A workaround is to use "Save As" to save the attachment, and then open the file manually.

Microsoft confirmed the issue yesterday afternoon in the advisory EX1528992, and early this morning began rolling out a fix to disable the "unnecessary" configuration. It expects that it'll take 12 hours for the fix to be rolled out, and users will have to close and reopen New Outlook to feel the benefit.

The problem raises some interesting questions. While we'd understand how something unexpected in Edge might cause rendering issues in New Outlook, losing the ability to open attachments because of something an Edge update did is concerning and may make some administrators reconsider the blast radius of software updates.

We asked Microsoft for more information about the "unnecessary configuration" and will update this piece should it respond.

However, the incident highlights connections within Windows that might surprise users. In addition, it shows that, despite a woeful showing on the desktop market share front - although Edge recently "surged" to almost 15 percent last month, according to Statcounter, up from around 10 percent for much of the year - Microsoft's browser can still capture user attention. It's just not the sort of attention the Windows maker might have hoped for.

It is also worth noting that Outlook Classic does not appear to be affected by the issue (according to Microsoft's EX1528992 advisory for Exchange Online).

Impacted users could therefore temporarily add opening attachments to the list of things the old software does that the new software doesn't. ®

Decentralized messaging platform Matrix is no longer the upstart it once was. The organization is now a serious challenger to the status quo as concerns about sovereignty and independence reach new heights.

At the organization's recent Malmo event, which grew attendance 30 percent year-on-year, the mood shifted from "yes, this is important" to "yes, we're rolling this out."

Matrix is an open standard for communication. It's a federated system, so there are no central hubs or overseas control to worry about, and it can be used for anything from messaging to voice over IP. Video is also supported.

The system is an alternative to services such as Microsoft Teams. It has become popular in the public sector as governments grow increasingly wary of relying on US tech giants for services that the Trump administration could switch off at will.

However, that increased interest and subsequent roll-outs have brought a level of scrutiny Matrix was not fully prepared for. A recent report regarding anonymous reaction to a roll-out of the technology by the European Commission took Element - the company that puts the lion share of the money into Matrix - by surprise.

One attendee at the Malmo conference observed, "First they [competitors] ignore you, then they laugh at you, then they fight you."

Matrix remains an open standard, however, the open source foundation behind it would certainly welcome more paying members.

"It's true that we're definitely getting more attention than we have had before," said Element CEO Matthew Hodgson.

"We are seeing Matrix ending up on the critical path of digital sovereignty as almost the poster child of the whole thing.

"I'm not sure that we necessarily do want to be the poster child," he added, "We didn't sign up to end up being ambassadors for digital sovereignty, roaming from government to government, but empirically that's where we've ended up, and yeah, that's going to attract attention from people who would rather everyone just kept in their traditional big tech systems."

This comes amid worries about the reliability of international partners, from early projects including France's Tchap messenger for the public sector and Germany's BundesMessenger through to more recent users, such as the Swedish public sector. Indeed, in one memorable demonstration during the Matrix event, messages were sent between French, Swedish, and NATO entities.

"I cannot overstate the strangeness," Hodgson added, "of having various national militaries phoning up every couple of days, saying, 'Look, we need to get on board Matrix too because we need to talk to this country and they're using Matrix, and we need to talk to that country, and they're using Matrix.'"

"It could be for defense, could be national security, could just be healthcare or education, public sector. But that is real. That is why there is a feeding frenzy."

Not that Element or Matrix is trying to build a better Teams than Teams. Patrick Alberts, Element's Chief Product Officer, said, "This is a digitally sovereign communication solution that, by and large, works like Teams in many aspects, but it's not set out to be the faster Porsche here..."

As for the reported complaints from the European Commission over Element's software implementation, Diana Gheorghe, EC head of sector for Communication and Collaboration Services, addressed them directly and described the anonymous complaints as "the natural friction of [a] high-profile rollout that we have done at scale."

Alexandru Vignide, EC service manager for Communication and Collaboration Services, who is very much at the pointy end of making the sovereign communication tools work, noted the speed of the rollout, which began with a proof of concept in Q4 2025, and hit laptops between July and September 2026.

The EC chose the Schengen Cloud Alliance (which includes OVHcloud) for hosting and, according to Vignide, "We wanted to provide a familiar experience for users." Not everyone is impressed with what is rolling out, but it's important to remember that change can be alarming, and this is still early days.

It is, however, a significant one. Gheorghe said that, as well as increasing resilience and business continuity, "We wanted to have vendor diversity. We also wanted to have data control and ownership of the platform."

Microsoft, as an example, would point to the sovereign capabilities of its system. As the Matrix conference kicked off, it announced new capabilities, including additional options for running AI, data services, and confidential workloads on customer-controlled platforms, and highlighted its sovereign partner ecosystem for customers who prefer local expertise.

However, Microsoft (and other tech giants) cannot escape concerns about who ultimately controls the data and what administrations might do with it. In 2025, a Microsoft representative memorably confirmed the US giant "cannot guarantee" data sovereignty to customers in France.

There is, however, a fly in the Matrix ointment. The platform is open source and, like so many open source projects, requires funding. Right now, Element accounts for much of the Matrix Foundation's funding. Hodgson finds this a little frustrating.

"If every government who is operationally dependent on Matrix put the smallest amount of money in the Foundation, the Foundation would have enough money to really start investing in the ecosystem and not be dependent on Element basically picking up the tab."

He acknowledged that things had improved in the last year, but said, "It's not good governance for an open standard to have one big player like Element effectively propping things up."

This isn't uncommon in the open source world, and it's a little disappointing given the ever-increasing importance placed on sovereignty in communications. Alberts commented: "There is confusion in the market that the thing somehow creates itself.

"No, it doesn't. It requires hardworking people. These hardworking people need to be paid, and that payment, in order to support government deployments going forward, ideally also comes from these governments. And some of them do a lot and do it in a great way, and that's amazing.

"But some don't." ®

Welcome back to PWNED, in which The Reg offers cautionary tales from folks who've left their doors wide open to attackers and lived to tell the tale. This week's misadventure involves unpatched older machines and a password that was supposed to send a message to crims.

Our story comes courtesy once again of leading red teamer and pen tester Dahvid Schloss of Suzu Labs.

A client asked Schloss to visit a factory and check the servers there for security holes. The company had been hacked six months prior, made some adjustments, and wanted to make sure that they were bulletproof now.

Schloss' mission, which he chose to accept, was to get domain admin rights.

Schloss drove over to the factory where he noticed a number of Windows 7 servers were connected to the CNC milling machines, even though Windows 10 was by then Microsoft's supported desktop OS. This was to be expected, as companies with proprietary software often use legacy OSes for fear of breaking apps.

At first, Schloss had a hard time breaking in because most of the servers were not connected to the domain. So there was no point in his running the EternalBlue zero-day that would have let him in.

"So I've been looking through these and most of them had been disconnected from the domain, so it's the right move," he said. "Like, yeah, you could run EternalBlue on the network, but you really weren't going to be able to get anywhere with it, cuz you get on the device and what you find: you dump LSASS [the Windows Local Security Authority Subsystem Service], you're like 'All right, there's local admin creds, but the local admin creds on those devices were different from the local admin creds on the majority of workstations."

Finally, Schloss went onto one of the servers that was connected to milling machines and found that it was connected to the network and on the domain. It was vulnerable to EternalBlue and he went and dumped LSASS and found a pair of creds from WDigest, one of which looked like an admin account.

But Schloss thought that it must not be a legitimate admin account when he read the password : "heyhackergof***yourself," but written in leet speak [l337 5p34k].

The resulting string was a strong password, but Schloss assumed that the admins were playing a game with him and left this credential there for him to find. Schloss assumed it wouldn't work and sat on it for a week.

During a briefing he had with the IT crew, he presented his findings and then asked who set up the admin account with this funny password.

"You could see everyone kind of looking at the senior manager of IT like 'that was you, wasn't it,'" Schloss explained. "He's kind of got this face. I'm like 'oh shit, like what?' Turns out that 'heyhackergof***yourself' was the actual password for his admin account."

On Monday, Schloss tried the username and password combo and, lo and behold, it gave him complete control of the entire network. He was then able to get access to other passwords and many of them were also insults directed at potential hackers.

Schloss understood that the team was still angry about a breach they had suffered six months prior and, should they be breached again, wanted to give the finger to the hacker who did it. Unfortunately, their system was still vulnerable because of the outdated machine that was hooked up to the domain and he was able to gain control.

The lesson here is really to patch your systems and avoid using insecure legacy OSes whenever possible, because it was an old machine that let Schloss in, not a weak password.

Indeed, "hackergof***yourself" in leet speak isn't a terrible password, just an embarrassing one if it's exposed. ®

HPE is giving its Alletra Storage MP X10000 another job: holding inference state that would otherwise be evicted from expensive GPU memory and recalculated later.

In testing validated by Kamiwaza using NVIDIA H200 GPUs, the Kamiwaza AI platform, and HPE Alletra MP X10000, the solution achieved up to 20× faster time to first token and 17× higher effective inference throughput versus running without KV cache offload.

The GPUs did not become cleverer. They had less repeat work to do.

The results move a storage decision directly into the discussion about AI economics. Model pricing gets plenty of attention. Paying expensive accelerators to process the same context repeatedly deserves some too.

Especially as enterprises move from chatbots answering individual questions to agents that can turn one request into a small committee's worth of activity.

Agentic workflows repeatedly process documents, system instructions and conversation histories. One agent retrieves information, another evaluates it, a third checks the result, and subsequent steps revisit material the system has already processed.

Some repetition is necessary. Recalculating everything is not.

KV cache retains intermediate attention state, allowing compatible requests to reuse earlier computation. Keeping that cache available can reduce the work required before a model starts producing its next answer.

GPU memory, unfortunately, is neither infinite nor priced as though it were.

Long contexts and concurrent sessions compete for that memory. When reusable cache is evicted and later needed, the inference engine may have to rebuild it. More GPU work, more waiting, and no additional insight to show for either.

HPE Alletra MP X10000 provides a shared tier for offloaded cache, while RDMA for S3 accelerates the storage data path. The proposition: retrieve reusable context rather than make the GPU calculate it again.

Compute once. Reuse often. An unremarkable principle, until the alternative appears on the infrastructure bill.

The Kamiwaza-validated testing used NVIDIA Nemotron 70B and more than 30 concurrent agents under large-context, multi-agent inference.

That matters because this is precisely the sort of real-world workload where repeated context and memory pressure can make cache reuse valuable. Several agents consulting overlapping material create a different opportunity from unrelated, short prompts with little context in common.

Those headline gains are workload-specific. Context length, concurrency, cache hit rates and transfer overhead determine the return. Twenty times faster is a test result, not a universal entitlement.

Buyers should examine the comparison as carefully as the multiplier: the same workload, with and without KV cache offload, and enough concurrency to represent production conditions.

The useful question is how much repeated computation their applications can avoid, and whether retrieving cached context is faster than rebuilding it.

A large performance number without its test conditions is mostly a typography decision.

Cache reuse addresses repeated computation. It cannot rescue a retrieval system that feeds the model half the company's document archive to answer one question.

HPE's broader pitch pairs the X10000 with Data Fabric for governed access to distributed data. The X10000's data intelligence node generates metadata, classification and embeddings at ingest to help applications retrieve relevant context.

That moves preparation closer to where data lands, rather than leaving every application team to assemble another downstream preparation pipeline.

The economic logic is straightforward. More precise retrieval can mean less irrelevant material in prompts and fewer attempts to obtain a useful answer.

Enterprise archives are rarely short of documents. Distinguishing the current policy from six superseded versions and a presentation called "FINAL_final_v7" is the more interesting problem.

Governance also matters when agents retrieve information and act on it. Relevant context still needs to be authorized context. Faster access to something an agent should never have seen is an awkward definition of progress.

That gives HPE two economic levers: improve what enters inference, then reuse computation where possible.

The distinction matters. Better retrieval can reduce prompt size and retries. KV cache reuse reduces processing work; it does not magically shrink generated answers or oblige an API supplier to lower its bill.

Cache offload is becoming a crowded market. HPE's strength rests on combining it with data preparation and governance, rather than winning another feature tick.

The practical test for organizations looking to get more value from their AI is how smoothly data preparation, governance and cache reuse operate as a single system.

Can applications find relevant, permitted information without unnecessary copying and preparation? Can the inference engine retain and retrieve reusable context efficiently? Does the combination sustain useful throughput as context windows and agent counts grow?

Those questions connect the architecture to observable outcomes.

Buyers should ask for cache hit rates, time to first token and sustained throughput on their workloads. They should also examine retrieval quality and whether the system completes the intended task. A faster wrong answer remains a wrong answer, however attractive the benchmark chart.

Cost per terabyte remains relevant. So does the cost of making a GPU read the same briefing for the seventh time.

For more information, visit the HPE Alletra Storage MP X10000 web page.

Contributed by HPE

Just two weeks after launching Gemini 4 Argon, Google has announced a better-performing version of the model that it will sell for the same price, at least temporarily.

Koray Kavukcuoglu, SVP at Google DeepMind and chief AI architect at Google, said Gemini 4 Carbon represents the company's strongest model yet, particularly in terms of cybersecurity and long-running tasks.

There's some irony in the name "Carbon", given the Chocolate Factory plans a carbon-dioxide-free future. Given Google's political timidity when it comes to names on maps, Gemini 4 Americium might be more representative of management's worldview. Gemini 4 Silicon is another missed opportunity — the chemical symbol is Si, a bobbleheaded nod to "Super Intelligence."

Priced initially at $2 per million input tokens and $10 per million output tokens, Gemini 4 Carbon sells for the same amount as its predecessor Argon during an indeterminate introductory period. When that period expires, Google will lift prices to $4 per million input tokens and $20 per million output tokens.

That's the same as Anthropic's Claude Opus 5.5, which Google clearly sees as Carbon's competition.

Last month, Google's ability to compete in the wake of the leadership change at its DeepMind AI subsidiary looked less certain.

But Carbon has posted impressive benchmark scores, taking the lead in SEC-bench Pro, ExploitGym, and CWE-bench, while landing at fourth place on WebDev Arena, behind Opus 5.5, GPT-6.1 Astra, and Sonnet 5.5. Carbon currently ranks second, behind Opus 5.5, in Arena.ai's leaderboard for Best Overall Agent.

Model ratings biz Artificial Analysis says Google's new model tops its newly formulated Artificial Analysis Intelligence Index v5 and comes in second to Claude Code Opus 5.5 on its Coding Agent Index.

Kavukcuoglu emphasized Carbon's cybersecurity capabilities, which will be available this week to an unspecified set of cybersecurity defenders and more eventually through the company's Fairwind access program. "On internal benchmarks, Carbon achieves more than 2x uplift in success rate for generating secure code over Argon," he wrote. "As another example, the Chrome team has reported that Carbon performed more than 10x better than another lab's large frontier model when generating high quality patches."

"Gemini 4 Argon and Carbon are now leading at the frontier," said Sundar Pichai, CEO of Google and Alphabet, in a statement [PDF] accompanying Alphabet's Q3 2026 earnings. "Overall, our first-party model APIs are now processing nearly 32 billion tokens per minute, more than four times what we processed a year ago."

While Google may hope to surpass Anthropic and OpenAI in terms of model performance and usage, it's already ahead in revenue. Pointing to the company's strong Q3 earnings, Emarketer principal analyst Nate Elliott in a statement told The Register, "In a sly nod to the way AI-only companies talk about their revenue, Google bragged about its half-trillion dollar run rate - almost ten times more than either OpenAI or Anthropic."

Google will release Carbon this week to select cybersecurity defenders. The company aims to make the AI model more broadly available "as soon as possible." ®

A scientist working on data gathered by the Rubin Observatory has found a strong candidate for our solar system's sixth known dwarf planet.

The Observatory is run by the USA's National Science Foundation and Department of Energy, who built it on a remote hilltop in Chile and equipped it with the 3200-megapixel Legacy Survey of Space and Time (LSST) Camera, which came online in June 2026.

As explained in a Tuesday announcement, Rubin Operations team member and project contributor Matthew Holman, of the Center for Astrophysics (CfA) at Harvard & Smithsonian, used search code he developed with CfA colleague Kevin Napier to analyze data captured before LSST operations commenced.

The Observatory says Holman's and Napier's code "is built to look at billions of detections of moving objects, compare them, and find the detections that correspond to the same object."

On October 13, the code spotted an object so bright and distant that it is likely unusually large.

"My first reaction was one of excited disbelief," Holman said. "Objects like this are very rare, so I immediately started checking whether I or the code had done something wrong."

Other astroboffins sprang into action, dug into other records, and soon found ten years' worth of relevant data.

Once they all checked their sums, the Minor Planet Center reviewed their work and announced the discovery of the romantically named 2026 ED138 and published lots of data about its orbit, which is at around 46 astronomical units (au). A single au is the distance from Earth to the Sun, and Pluto's orbit averages 40 au.

Rubin Observatory revealed its part in the effort and declared that 2026 ED138 is a good candidate for certification as a dwarf planet.

The 2026 International Astronomical Union defines a dwarf planet as "an object in orbit around the Sun that is large enough to pull itself into a nearly round shape but has not been able to clear its orbit of debris."

The Union's definition of a full planet includes having "an important influence on the orbital stability of the other objects in its neighbourhood." Dwarf planets whirl about the solar system in relatively close company with many rocky and/or icy objects that orbit under the influence of other bodies, not the nearby dwarf planet.

Rubin boffins think ED138 is between 600 and 1500 kilometers across. If the lower number is correct, it's the smallest dwarf planet yet observed. Should the larger number be correct, it would be smaller than Pluto, Eris, and Haumea, but larger than Makemake and Ceres.

Scientists at the observatory think ED138 has evaded detection to date because it was found in a region of sky close to the plane of the Milky Way — meaning observers had to pick it out from light cast by billions of stars.

"Rubin Observatory's regular and continuous monitoring of the entire Southern Hemisphere sky, combined with powerful algorithms that sift through millions to billions of points of light to identify those corresponding to small moving objects, enabled the team to spot ED138 within the crowded field," the facility boasted. ®

Japan's Fujifilm has decided the tape market has so much upside that it needs to boost production of its Linear Tape-Open (LTO) products by five times.

LTO is the sole surviving non-proprietary tape format and is kept alive by HPE, IBM, and Quantum Corp, who head the LTO Program.

As The Register recently reported, shipments of LTO cartridges went backwards during 2025, when buyers shelled out for 160.3 exabytes worth of LTO compared to the record 176.5 exabytes; worth of tape cartridges sold in 2024.

The LTO program attributed the 2025 result to difficult macro-economic conditions, noted that sales picked up strongly in the second half of the year, and predicted record sales in 2026.

Fujifilm seems to agree with that optimistic outlook, as it today announced a plan to spend ¥10 billion ($64 million) over the next two years to expand a tape plant in the Japanese city of Odawara.

The company said AI has made this a sensible investment.

"The rapid spread of generative AI and cloud services is driving exponential growth in the volume of data generated worldwide," the company's statement reads. "As a result, establishing environments in which the enormous amounts of data accumulated through AI training and inference as well as the digitalization of business activities can be stored and utilized securely and efficiently is becoming increasingly important in supporting the digital society."

Fujifilm says its investment will go towards building more tenth-gen LTO tapes, which come in variants that boast native storage capacity of either 30TB or 40TB, and transfer data at 400 MB/s. Many LTO users employ data compression to more than double the capacity of each tape cartridge. Decompressing data takes time, and shifting data off tape doesn't happen in a hurry. Tape users, however, nearly always use it for offline or nearline archives, and understand that access times are not the medium's strong point.

The LTO Program has already defined four more generations of its format and delivers updates about every four years. LTO 11 is therefore due in about 2029 and will reach 70TB capacity. Version 12 will climb to 120TB, before unlucky version 13 shoots up to 210TB. Version 14 will jump to 365TB per cartridge!

The LTO program used to insist that tape drives could read the current past generations of cartridges. That ended with LTO-10 devices, which can only read the current generation of tapes. ®

Shares in Google's parent company Alphabet dived by almost ten percent on Wednesday, after the web giant admitted AI-powered searches earn it less money than it earns by serving ads with its old-school search engine.

Alphabet on Wednesday reported its Q3 results, which included news that its year-over-year search query growth rate is the highest it's been in the last five years. Some of that growth comes from users adopting "AI Mode" - Google's latest LLM-powered search system.

Google's revenue from search rose 12 percent year-over-year to $63.25 billion, a little lower than the 14.5 percent growth the company reported in last year's Q3.

During Alphabet's earnings call, chief business officer Philipp Schindler said the slowdown was caused by "lower monetization we were seeing in AI Mode."

That's a worry given that Google is spending enormous sums on infrastructure. $51.4 billion went out the door in Q3, around two thirds of it going towards servers and networking equipment. The rest funded datacenter builds.

That spending appears to have impacted profitability, as Alphabet reported $26.17 billion net income for the quarter. Last year's Q3, during which Google revenue was $102.35 billion, produced $35 billion net income.

Schindler argued that AI search will pay off.

"On AI Mode, many of these queries are upper funnel queries, allowing us to capture new, high-value commercial intent," he told investors. "In addition, the majority of AI Mode queries are incremental, creating more monetization opportunities over time."

Google tried to seize one of those opportunities today, by turning on ads alongside AI Mode searches around the world, after first testing them in the US only.

But for now, AI Mode searches aren't as lucrative as Google's existing search business, which brought in just under half of Alphabet's $125.8 billion quarterly revenue.

Google's cloud is its next-best earner, at $24.75 billion for the quarter, but it has troubles too because the Big G doesn't have enough capacity to serve all its customers. The company has started to offload some customer workloads to third parties, and warned that will dent its future margins.

Thomas Kurian, Google Cloud's CEO, said the company has over 100 companies using its TPU silicon, adding that "All of the inventory for TPU systems are associated with committed contracts" - meaning Google can see some revenue to cover some capex. The company expects to book revenue from TPU sales to third parties in coming quarters.

CEO Sundar Pichai was full of optimism for future AI revenue, noting that Google's Gemini app has won a billion users faster than any other product in the company's history, and praising the performance of its models. He said Google is moving fast to create agentic AI tools, but will focus on business customers for starters, "because the bar for security, safety, and governance is extremely high."

"As a next step, we are bringing these features to Search and also, of course, to our billion users on the Gemini app," he added.

Pichai didn't say when agentic AI will come to search or if he expects it will avoid the monetization issues AI Mode has already caused.

Yet the company also stuck to its $195-$205 billion capital expenditure forecast for this year and said it intends to spend "significantly" more in 2027. ®

Google is working with NASA to develop a robotic medical assistant that could help astronauts on future deep-space missions.

Chocolate Factory's public sector VP Jim Kelly said in a Tuesday blog post that the company's latest moonshot could take its technology far beyond the lunar surface, with Google's public sector arm partnering with NASA's Human Research Program, Johnson Space Center's robotics team, and the Jet Propulsion Laboratory to develop a robot that could assist astronauts with medical tasks, cargo handling, and routine maintenance.

The robotics project builds on the Crew Medical Officer Digital Assistant (CMO-DA), an AI-powered medical decision-support system Google has been working on with NASA since last year. CMO-DA, which is powered by the Red Hat-supported open source local AI tool RamaLama, is already being evaluated on Earth-bound copies of computers used aboard the ISS. It's designed to run AI models locally for evaluation of medical issues, which will be a key part of future space missions due to the communications lag between Earth and astronauts in deep space.

"An astronaut on a future deep space mission to Mars could face severe operational challenges—including round-trip communication latencies of up to 44 minutes, making real-time medical guidance for crew nearly impossible," Kelly said in this week's blog post.

Google and NASA envision the robot as an assistant to a human crew medical officer, helping with medical tasks and doing other work when bedside assistance isn't required.

"To maximize its utility for a lean crew, the robotic system is designed as a versatile 'payload' handler capable of managing cargo, logistics, and equipment," Kelly explained. "When not needed for routine maintenance like organizing tools or securing compartments, it can pivot to medical support."

As one example of its potential use, the Google public sector VP explained that, were someone to experience severe back pain potentially caused by kidney stones, the robot could fetch and prep an ultrasound machine, leaving the crew medical officer to interpret the scan and decide on treatment.

"The ultimate goal is to build a robotic system that can execute a variety of tasks via simple voice instructions, under the astronauts' direct supervision and command," Kelly said. "For instance, the robot might be directed to assist with routine health monitoring—such as capturing basic vitals or simple ultrasounds—under the strict guidance of a human crew medical officer."

So, not a full-fledged digital medical officer as much as an extra set of hands for an overworked human doc.

As for the status of the project, it doesn't appear to have advanced too far yet. Google outlines several technologies being explored, including JPL's work adapting satellite situational-awareness tools originally used to track objects in orbit to help the robot navigate 3D spaces. The project remains at the proof-of-concept stage, with early laboratory demonstrations conducted, but Google hasn't confirmed whether a fully integrated, space-ready robot exists.

Kelly explained that Google used DeepMind's MuJoCo physics simulator to test the robot's movements, and that the system scored an initial 85 percent success rate in early lab tests involving locating and moving a medical bag - not exactly proof that it's ready to assist with medical procedures, but it's a start.

Google referred us to NASA for questions, but the space agency declined to get into specifics. The agency only told us it's working on a variety of "super intelligence" (eye roll - that's AI to those of us operating in reality) projects for its many missions, and that it's working with partners to assess them for future applications. ®

Why ask a human to complete a coding test when they can just use an AI to help them? Anthropic engineering manager Felix Rieseberg says he has stopped using traditional coding interviews for his team.

"Being able to quickly define problems and find elegant solutions was a weak but useful predictor for how successful you'd be on my teams," Rieseberg explained in a blog post on Tuesday. "Small coding challenges in real code did tell us something meaningful about candidates."

No longer, he says. The coding interview is dead, as far as he's concerned.

"There's no more point in giving you a little React challenge than there is in quizzing people about assembly; Claude is now better at both," he insisted. "Our most impactful builders are successful regardless of how good they are at these tests."

In a response posted to social media, Jaana Dogan, principal engineer at Google, said much the same. "I'm not doing a ton of coding interviews nowadays but I'd be interested to interview how people use coding agents," she wrote. "That itself gives me more signals about candidates' coding skills."

Several months ago, Google reportedly began reevaluating how it conducts coding interviews and what it looks for in candidates. And while it would be surprising if similar conversations haven't already occurred at other large technology companies where AI agents have demonstrated their coding capabilities, the rest of the tech industry appears to be moving more cautiously than the AI enthusiasts.

"We are not seeing tech companies abandon skills assessments widely across the industry—at least not yet," said Nicole Bachaud, labor economist at ZipRecruiter, in an email to The Register. "In fact, based on our October 2026 Employer Survey of 1,500 employers, we see the tech industry (Information sector specifically) is actually the most likely to test for skills, with 62.5 percent of employers requiring formal skill tests on initial applications."

Coding interviews at Microsoft, and later Google, Amazon, Facebook, and other tech giants, vary, depending on what the hiring organization aims to evaluate. They may probe technical knowledge related to a specific subject, coding ability in a particular language, general problem-solving skills, team-oriented behavior, or other skills.

Stripe's software engineer interview process, according to Educative, consists of a phone screening stage, a technical phone interview, and several further on-site interviews.

Rieseberg actually worked there and describes how founder Patrick Collison urged new hires to pay attention to things that seemed out of place. "His bet was that newcomers would spot friction and opportunities that old-timers had long stopped noticing," he said.

And Rieseberg appears to have absorbed that viewpoint based on his dismissal of experience: "Hunger beats tenure," he said, perhaps unaware that technology companies have gotten into legal trouble for jettisoning costly veteran employees so they can be replaced by hungry, less expensive young professionals.

Technical interviews are seldom just about technical mastery. There are after all many other ways that employees can add value to an organization beyond testable computer science skills. And when employees have access to AI models that can answer technical questions better than job candidates, it makes sense to look more closely at the creativity and social skills people bring to the table.

"The change at Anthropic might be reflecting a shift away from technical skills towards more human, soft skills," said Bachaud. "As AI becomes fully capable of writing code, tech employers may be pivoting away from rote technical testing and heavily toward human-centric soft skills."

She argues that while AI may be able to code well, it can't replace critical thinking. "A staggering 61.5 percent of Information employers now report that 'critical thinking' is a primary skill lacking in today's candidate pool," she said.

"I don't think the coding interview is going to come back," said Rieseberg. "It's easy to get sentimental given how many copies of 'Algorithms and Data Structures' were bought in hopes of unlocking good jobs and a better future — or how the death of the coding interview is another milestone in software engineering reaching a new level of abstraction."

At the same time, he's not certain what will replace it.

"I'm probably not the only manager trying to figure out how to interview for people who build well with AI," Rieseberg wrote. "I haven't quite gotten there yet — but I'm scanning resumes and portfolios for little prototypes that exhibit taste, empathy, and principled thinking. If you have figured out something better, I'd love to hear how."

Bachaud sees technical evaluations changing instead of going away, noting that 70 percent of Information sector employers have switched to proctored or in-person skill assessments as a way to deter cheating with AI.

"Ultimately, tech companies aren't necessarily testing less, but the way they are testing is changing," she said. "As AI automates a lot of manual and technical components of jobs, work skills may become more universal. That could lead to an increase in job switching and turnover once economic conditions improve. In our recent survey of workers, we found that AI-related work concerns are causing a quarter of workers to start thinking about their next role. And as assessments and skills are changing, what roles are available to them might look more different than ever before." ®