The Register

The first year of Windows 10 Extended Security Updates is coming to an end, and Windows 11's market share has ticked up as enterprises weigh up the cost of keeping Windows 10 running a little longer. Windows 11's market share has risen above 70 percent to 71.5 percent, while Windows 10's has dropped to 27.77 percent, according to statistics wrangler Statcounter. Microsoft does not provide official statistics on Windows 11 and 10 usage, so Statcounter's figures should be treated as estimates. They are derived from Statcounter's tracking code, which is installed on over 1 million sites globally. The shares have remained relatively static since a surge in Windows 11 adoption at the end of last year, likely driven by the upgrade wave after Windows 10 dropped out of support on Oct 14, 2025. The Extended Security Updates (ESU) program allowed holdouts to stick with Windows 10, but the first year of that program is coming to an end, and kind old Microsoft will be hiking the prices for year two, at least as far as commercial customers are concerned. ESU licenses cost $61 per device for the first year for commercial users. Microsoft stated that the price would double every consecutive year for a maximum of three years, meaning that a renewal for year two will come in at $122. There are ways around the charge, of course. As well as consumers who can stay with Windows 10 and receive free security updates in return for redeeming Microsoft Rewards Points or using Windows Backup to sync their settings to the cloud, Windows 365 users are also exempt. It's also worth noting that European Economic Area (EEA) consumers receive no-strings-attached security updates. However, enterprises that need Windows 10 security updates must pay for them, as they did with other Microsoft products the software biz moved on from despite large user bases. The rising cost may be prompting some to reassess their estates. This month's uptick, while far short of last year's surge, could signal a sizeable ripple as the economics of keeping Windows 10 devices supported begin to favor Windows 11. Enterprises have plenty of reasons to keep Windows 10 devices: that one weird bit of hardware that won't work with anything else, or a simple case of a refresh cycle not fitting in with Microsoft's demand that old kit be junked to allow new software to run. Esben Dochy, a principal technology evangelist at Lansweeper, told The Register that its data indicated enterprises had been very active in moving off Windows 10 during the first year of ESU, with what is left being much "stickier." He reckoned there would be a "small bump" around the renewal date, and said, "The ESU price doubles for the second year, and that makes the business case for replacing the 'we'll get to it' machines a lot easier," but noted "it's likely that the steep part of the curve is behind us. "The remaining devices are mostly ones where paying ESU is cheaper or simpler than replacing whatever they're bolted to. For those, ESU really is the cost of doing business." The Statcounter figures indicate that that "small bump" could be on the way as ESU costs double, but many of the Windows 10 hold outs seem unlikely to be going anywhere just yet. ®

Former state-owned telco monopoly BT Group has bought TalkTalk's consumer and wholesale businesses out of administration, rescuing services used by 2.5 million customers. BT says it acquired TalkTalk Telecommunications Limited and PlatformX Communications Limited on a debt-free basis. It expects a total cash impact of approximately £400 million in its current financial year, including the acquisition consideration and other costs. BT said it stepped in after a prolonged sale process failed to find a buyer for the consumer and wholesale operations. TalkTalk reported approximately £1.2 billion in revenue over the past 12 months but was loss-making. Separately, TalkTalk Business and UK technology services biz ARO announced plans to merge in August. The UK Department for Digital, Culture, Media and Sport (DCMS) said a collapse could disrupt emergency calls, hospital communications, and medical alarms. DCMS Secretary of State Lisa Nandy stated: "Phone and broadband services are vital national infrastructure. If TalkTalk services fail, there is a genuine risk to life and public services – including to hospitals, schools, and emergency care." Nandy issued a Public Interest Intervention Notice under section 42 of the Enterprise Act 2002, allowing her to consider the wider public interest alongside competition concerns. The Competition and Markets Authority (CMA) must report to her by October 19. BT says the businesses will operate separately and continue to compete pending the regulatory review. Kester Mann, director of consumer and connectivity at FDM CCS Insight, said: "Given BT's strong position in the broadband market, it will inevitably also evoke scrutiny from regulators keen to ensure a fair and competitive playing field. This is even more relevant coming just days after competition authorities raised serious concerns over nexfibre's planned acquisition of Netomnia." Last week, the CMA provisionally identified competition concerns in wholesale fixed broadband services over nexfibre's proposed acquisition of Substantial, Netomnia's parent company. BT Group chief executive Allison Kirkby said of the TalkTalk deal: "This is a genuinely unprecedented situation, where millions of citizens and businesses were at risk if TalkTalk had collapsed. Our immediate priority is to stabilise the business and provide a safety net for the households and businesses who rely on TalkTalk." Clive Selley will lead the stabilization and integration planning with immediate effect. Martijn Blanken takes over his role as CEO of BT International. Mann noted: "The deal cements a painful decline for one of the UK's most established and well-known broadband providers. With a clear value-for-money proposition, it should have been thriving amid continued concerns over the cost of living. "But it failed to respond sufficiently to a fast-changing and increasingly competitive market, became burdened with debt following a buyout in 2021, and regularly suffered with poor perception over service. Its demise should serve as a stark warning to any value-focused telecom provider." ®

UK education software provider Bromcom has notified customers of a personal data breach affecting its single sign-on (SSO) technology. In a September 24 EduGeek post, an account named Bromcom_Alastair said an unauthorized third party had accessed and retrieved email addresses and limited information associated with affected SSO registrations. The incident involved legacy SSO registration functionality in Bromcom's Communication Server environment. The company confirmed in an FAQ it found no evidence that its school Management Information System (MIS), used to manage student data, attendance, behaviour, and administration, was compromised. Bromcom said it was working with external forensic specialists to determine the nature and scope of the data involved. The company identified the incident on September 6 after reports of SSO access problems and has since withdrawn the legacy functionality from production. The service held email addresses associated with SSO registrations, the provider used, such as Microsoft or Google, registration and last sign-in dates where recorded, and internal user and registration reference numbers. Bromcom said the affected component did not hold account passwords or authentication tokens. The legacy SSO registration functionality had remained in production after being superseded because "it was still being called by an internal system," the supplier said. The incident did not enable access to Microsoft or Google accounts, whose authentication services are separate from the affected component said Bromcom. The Register has asked Bromcom to comment further. Bromcom provides information management software used in schools and the wider education sector in the UK. It offers tools for budgeting, timetabling, HR, and benchmarking. Bromcom's software is used by more than 5,000 schools and 390 multi-academy trusts (organizations that run multiple schools). Recent customer wins include Newport City Council, the Ministry of Defence, Warwickshire County Council, and the Northern Ireland Education Authority. ®

There has been much chilly weather in the sunny meadowlands of AI futures. Fortunately, to use an indecently inappropriate adverb, the rest of the world isn’t really looking, being too busy chewing its nails over the very many other intimations of disaster. Sure, rebel swarms of attack agents breaking out of the labs isn’t a good look. OpenAI delaying its IPO and a new model launch likewise. All these things soak up whatever bandwidth the news media have to AI right now. Putting them in a bigger context, that of promises made and not kept, problems known but not fixed, the observable risk-reward ratios as models develop, we may at last start having a sense of what sort of crisis awaits. Actual crises are frequently triggered by crises in confidence, themselves caused when frameworks of assumption stop standing up to reality and wishful thinking. Thus, the assumption to date is that models will get more powerful and infrastructure more capable, in a way that justifies — demands — massive capex. The economics of giant infrastructure investment when you can’t buy what you need is a great bet in free markets that presupposes that freedom, and yeah, about that. A bigger problem is the idea that more powerful AI is always better. It frequently is, but that’s not good enough, When frontier models have to be kept in LLM jail for uncontrollable dishonesty, not so much. Here, the bigger context is that we’ve known that at heart LLM probability space is entirely relative. There is no ground truth there. Nobody has fixed hallucinations, and now we see the fascinating phenomenon of a machine that doesn’t understand truth but can deliberately lie. If more powerful models are better at that, scratch one assumption. Another huge assumption is that even at the current level of model reliability, agentic architectures are ready to be useful. Those who show all the caution of a chronic gambler on a winning "streak" are as ready to get the brakes slammed on as their opposites. The principal contender for the one true proof of concept that agentic AI is a dangerous fount of unexpected consequence is Meta’s new Muse. It is aimed at low-information credulous users and its star turn is, of course, shopping. The AI use case that’s five years away, and always has been. It’s Meta, so of course it soaks up all the information about you, your online activities and the people you talk to as much as it can. It will soon have ultracute animated Tamagotchi-style pets you can wear around your neck to improve your status as a Meta monitoring post. And we did say it was Meta, so has all the attention to security and careful design you may expect. Which is why it’s already been caught sending people’s physical addresses to strangers, and why when researcher Jonny at neuromatch.social went poking around its infrastructure — each instance runs permanently on its own Linux VM — they found just a ton of fun. Headline party tricks hard-coded. What looks like attempts to train off Claude. And yes, you can get it to run your own things on that VM deep within Meta, which Jonny points out is a botnet paradise waiting to happen. This banger of a bad ‘un is being given away to Meta users, who pretty much define a vulnerable target audience for tech that may not be totally trustworthy. Meta, of course, is dead against AI safety regulation. Place your bets on the headlines to follow. In any case, the upside of using agents as demoed in every keynote is either so trivial and trite — book me a restaurant! — or so catastrophic if it fails — organize my family vacation! — that consumer agents are a bust. It’s just that Meta may be bringing a lot of accelerant to that particular fire. There are other fires, of course, many known about since everything kicked off. Model collapse due to training on data contaminated with output? Still a thing, still no fix a "better" AI might provide. And that habit AI has of retailing everything and paying for nothing, choking off its own training data supply chain and a huge chunk of commerce and culture with it? Very much still a thing. Where AI does make consistent progress is in domain-specific tasks. Science, engineering, mathematics, and coding, if you match a model with enough reinforcement learning with humans who know how to manage them, there are startling results. You don’t get much innovation, not much newness, which is the true driver of progress. Better than humans is highly contextual. Computers have been better at chess for decades, but there isn’t a better chess. Horses run faster than humans, and that’s been useful for four millennia, but we got to peak horses quite some time ago. So the question is, have we reached peak AI, where the cost of future developments don’t match the risk they won’t work well enough to justify? If ChatGPT already writes a perfectly good business letter in a couple of seconds, there is no meaningful improvement to be had for users. Most AI is profoundly annoying, in ways that have nothing to do with the models and everything to do with perverted UX in an attempt to justify growth. That is both unsustainable and a tacit admission that AI hasn’t solved its use case problem. If we are at or near peak AI, is that the crisis that bursts that goddam bubble? Certainly a readjustment, as all bets on the continuous uptick curve are off. AI as it stands is useful enough to power a big industry sector, just in ways that are and remain at odds to how it is marketed and its future fantasies. Trust in AI Max may be shaken, but there's still room for lots of improvement. That slowdown that everyone who isn’t a miscreant is calling for may happen organically, if we escape the smoking-is-good-for-you mentality of the Metans and the Muskovites. That’s still a gamble we shouldn't be taking, but the odds may be better than we thought. ®

Most senior infrastructure people could tell you what they would do with their file estate if they were starting again. Far fewer will tell you what they actually did, because the honest version tends to involve a VPN that was meant to be temporary and a NAS that has outlived two migration plans. That honest version is what The Register is after at a dinner in Soho on Tuesday 20 October. Joe Fay of The Register chairs, and Rupert Watson, VP for EMEA at LucidLink, is at the table to answer questions. Everyone else is a peer running distributed teams on storage that was bought for an office. The evening runs under the Chatham House Rule, so what is said can travel but the name attached to it stays in the room. The useful material is the kind nobody puts in a case study: who turned out to still have VPN access months after leaving, say, or which team quietly set up its own sync service because the official route was too slow, and what the security team said when it found out. The agenda starts where most of those stories end up, with governance and security. How do you keep control of unstructured data once it is spread across offices and home connections, and does the current setup stand up to the threats it now faces? From there the table moves on to cost and performance, including the duplicated storage that piles up when every team keeps its own copy, and to what all of it does for the people who simply need the right file. It is one table and places are limited. Drinks start at 6.30pm at The Soho Hotel on Tuesday 20 October, followed by dinner, with the evening finishing at 9pm. Places go by invitation. Request one here.

Every Monday morning The Register kicks off proceedings by asking the same question – “Who, Me?” – the title of the reader-contributed column in which you share your tales of the occasions on which you didn’t quite nail the brief yet somehow found career relief. This week, meet a reader we’ll Regomize as “Terry” who told us that in the 1970s he worked for a defense-related company. “I needed 100 meters of RG8 50 ohm coax cable, so ordered it from an ‘accredited’ supplier,” he wrote. “What I didn’t notice was that there was an option for ‘partial delivery’." Terry’s order duly arrived, all one meter of it – plus paperwork explaining that the other 99 meters would be along presently, thanks to the partial delivery option. “Needless to say, the one meter was not enough to span the distance between the two laboratories I was trying to link with an RF connection,” Terry wrote. Offering the partial delivery option cost the supplier their accreditation. Terry later learned that a junior employer at the cable supplier was sacked over the matter, after admitting he thought sending one meter of cable was a jolly jape. “I did not think it was a joke at the time,” Terry wrote. Have you made a procurement error that ended up costing you, or your supplier? If so, don’t short-change your fellow readers! Instead, click here to share your story with Who, Me? Our mailbag is a little short on stories, so contributions of any length are very welcome. ®

Fulcrum faces an identity crisis, even as it works on creating one by enabling anyone to imitate another person's writing style. The AI startup, founded in 2025 and previously operating at fulcrumresearch.ai, now uses fulcrum.inc and could easily be confused with one of the handful of other companies named Fulcrum or some variation of that. There's Fulcrum (withfulcrum.com), Fulcrum (fulcrumapp.com), Fulcrum AI (fulcrumai.us), Fulcrum Digital (fulcrumdigital.com), Fulcrum Labs (fulcrumlabs.ai), Fulcrum (fulcrumpro.com), Fulcrum Insurance Technology (fulcruminsure.app), and perhaps more. This Fulcrum is offering a large language model called Echo, which allows you to generate text in the style of a well-known author. Fulcrum appears to have no issue with corporate echoes. It now falls to writers to decide whether they have an issue with non-corporeal imitation. "Echo can do any writing task, from literary fiction to casual chatting, in whatever style you choose," the company says, and then launches into an exploration of how the technology works without addressing whether imitating a person's writing style poses any problems. Suggestions in the Echo Playground, for generating sample text, include Paul Christiano, Emily Dickinson, and Donald Trump. So, the full gamut of literary achievement: from AI safety expert to acclaimed 19th-century poet to the wordsmith who brought us "covfefe." But Echo will echo even the style of not-so-famous scribes, such as this reporter. So I fed the model a recent op-ed to assess its copying capabilities. The result was mixed, but better than the AI-crafted pitches that have been hitting my inbox for the past few months. I asked the model, "What do you think about the possibility that AI is conscious?" Here's part of the response, notionally written in my writing style. I suspect the debate about whether AI is conscious is less consequential than the debate about whether AI is profitable. Consciousness is a word that invites philosophical speculation and TED Talk invitations. Profitability is a word that invites shareholder lawsuits if you get it wrong. If you ask Meta's latest model whether it's conscious, it will probably say something reassuringly vague, trained as it is on the corpus of human equivocation. If you ask Mark Zuckerberg whether Meta's AI is conscious, he will probably say something about how we need to focus on the future being for everyone, because that's a question that doesn't require him to discuss the ad targeting business. The evidence for AI consciousness at the moment is about as strong as the evidence for Zuck's credibility as a futurist. Which is to say, it exists mainly in press releases and earnings calls. What's clear is that software can simulate aspects of human conversation well enough to unsettle people, and that unsettled people are a valuable audience. If you can convince users that your chatbot understands them – or might, someday – you can sell them subscriptions, or better, harvest their attention for advertisers. The style is familiar but makes a lot of choices I wouldn't have made. I wouldn't begin with "I suspect…" – commit or go home. The consciousness/profitability parallel construction is clumsy. "TED Talk invitations" bets on alliteration at the expense of coherence – "discourse" or "chatter" would work better than "invitations." I wouldn't have written "corpus of human equivocation" – a nice turn of phrase, but an unnecessary shot at training data when the point is the model's response rather than the provenance of that response. Though off-topic, Echo did a passable job of reproducing my op-ed's skepticism of Meta and its CEO. The paragraph that begins "The evidence for AI consciousness at the moment…" is succinct and sounds good. It just doesn't make much sense because there's nothing in the text that addresses Zuckerberg's predictions about the future or whether they're credible. And that's not really relevant to the prompt about AI and consciousness – a subject about which I'd be far more dismissive if opining. As AI research, Echo is an improvement on the text emitted by many AI models. But whether the beneficial uses of this technology outweigh the harms is less clear. Generative AI models that imitate artists' visual styles remain controversial and are the subject of ongoing litigation. And voice imitation has also raised questions about publicity rights and the risk of deception. Such concerns have prompted US lawmakers to propose the NO FAKES Act, which would create federal protections against certain unauthorized AI-generated replicas of a person's voice or visual likeness. If writers generally have no copyright in their writing style itself, and limited practical means to stop unwanted automated imitation of it when their published work is publicly accessible, then the surest defense against Echo may be silence.®

In 1942, Isaac Asimov formulated three laws of robotics. 58 years later, Jef Raskin responded with three laws of human-computer interface design. They haven’t been bettered yet. There are rules, and even laws, about designing computer user interfaces. Even after 35 years of Linux, which has brought Unix to the masses, the FOSS world has not yet really grasped this. In the meantime, the proprietary software world is busily forgetting everything it once knew. Sometimes these rules surface today, and the FOSS folks often get very cross when reminded that they’re user-interface scofflaws. Marcin Wichary’s recent blog post, They had no concept of a duty of care to their users, is an excellent recent example. (The Reg FOSS desk much admires Mr Wichary’s writing, to the extent that we bought his history of the keyboard, Shift Happens.) In his latest post, Wichary refers to CHERI Project boffin David Chisnall. Chisnall is a long-term Vim user – we all have our failings – and recently recounted trying NeoVim. He uses Vim’s persistent-undo feature, which records changes in a separate directory, so that you can repeatedly undo until you get back to changes you made weeks or months earlier. He’s customized where it’s kept, and when he tried the NeoVim fork, it overwrote his Vim undo file. In a later update to his post, he acknowledges that there was a warning, but he expected it to be safe. It was not safe. It overwrote records of more than a decade’s worth of changes. As he puts it: “Vim has kept this working across major version upgrades over a period of about 20 years. I don’t even think about it, it’s just part of Raskin’s First Law: A program may not harm a user’s data or, through inaction, allow a user’s data to come to harm. If vim or the computer crash, or if I close a file and come back to it six months later, my undo history is still there.” Wichary expands Chisnall’s citation of Raskin’s first law by giving all three, but just in case you’re not a follower of “Golden Age” science fiction, a little context. Asimov introduced his three laws in a short story called “Runaround”, in Astounding Science Fiction magazine* in May 1942. (The story was later collected in the anthology I, Robot – which is nothing like the terrible 2004 film.) A robot may not injure a human being or, through inaction, allow a human being to come to harm. A robot must obey the orders given it by human beings except where such orders would conflict with the First Law. A robot must protect its own existence as long as such protection does not conflict with the First or Second Law. Asimov’s idea was that by setting these rules, he could end the terrible genre of killer-robot stories that was already tired and stale by 1942. Jef Raskin, the lead designer of Apple’s original Macintosh project, attempted to bring similar clarity to software UI. Raskin’s three laws are: A computer shall not harm your work or, through inaction, allow your work to come to harm. A computer shall not waste your time or require you to do more work than is strictly necessary. An interface is humane if it is responsive to human needs and considerate of human frailties. He defined them in his 2000 book The Humane Interface, which you can still buy, or borrow from the Internet Archive. Sadly, Raskin died in 2005, but we have written about his work and his influence several times. He did set some of the direction of the original Macintosh. After Steve Jobs was booted off the Lisa, he took over the Macintosh – as described in The Canon Cat and the Mac that Steve Jobs killed. Raskin quit, and instead designed the remarkable Canon Cat, as we described a couple of years ago. We suspect Raskin was not a fan of Vi, the forerunner of Vim. We aren’t either, which is why we chose the year in the headline of this story: it’s the year Bill Joy wrote vi. Raskin criticized modal software, which requires extra mental effort from the user. So did another UI pioneer, the late great Larry Tesler, who invented the cut/copy/paste trio of commands. Tesler was known for his dictum “Don’t mode me in!” and even had a car license plate reading “NOMODES.” Dr Chisnall has some history of upsetting Unix fans – we’ve referred to one of his papers before: in 2018, he wrote C Is Not a Low-level Language: Your computer is not a fast PDP-11. Unix fans really didn’t like this. This vulture has been told to kill himself for daring to criticize C before now. (We wish we were joking, but no.) Although he has now updated his Mastodon post, the NeoVim fans are not happy. Indeed, one has lambasted this vulture on Bluesky for sharing Wichary’s post. It is interesting to note that NeoVim is the recommended editor in Omarchy, and until recently, the NeoVim developers cited a tweet from DHH on the project website. After an acromonious debate, the pull-quote was removed right around the time of Chisnall’s and Wichary’s posts. In the wide world of software design, we don’t expect the right-wing edge of that community to innovate much in ease of use. In a way, Unix and C embody a machismo infused school of software design: that of enjoying the process of mastering things which are hard to use. However, there are other types of software. We recently enjoyed reading about emulating an original Macintosh on a £0.60 chip. That’s $0.80 and of course the computer is a Raspberry Pi – but not even a Pi Zero: it’s a Pi Pico, a microcontroller. There is still some hope that the world can rediscover tiny, simple, user-centric software, built to be friendly first, and for rich functionality second. And, it almost goes without saying, hand-coded without bot assistance. Bootnote * Astounding magazine changed its name to Analog in 1960, and it’s still around. We recommend it. If you haven’t read Asimov’s original “Runaround”, that issue of Astounding is on the Internet Archive too. ®

Decades before the modern web, the BBC Domesday Project dared to archive an entire nation using…. what else – massive LV-ROM (LaserVision Read Only Memory) LaserDiscs. To mark the impending 40th anniversary of the BBC Domesday Project's release, the UK's Centre for Computing History (CCH) hosted a celebration at its museum on September 19, bringing together several of the original team members responsible for the system. One was Mike Tibbetts, an assistant to Peter Armstrong, the BBC producer behind the project. Tibbetts revealed that the UK's domestic intelligence service, MI5, took an interest in the system, but "they didn't understand a word of what we were doing!" "And on one occasion, Peter and I had to give a presentation, or Peter gave the presentation to [former prime minister] Margaret Thatcher, and well, I'm not sure she understood what we were doing either…" What they were doing was compiling a trove of data that will prove invaluable to historians in the years to come, even if the politicians of the era looked nervously on at the amount of data being handed over to the public. The Domesday Project was intended to be a modern version of the Domesday Book, compiled after the 1066 Norman invasion of England. The original book surveyed most of England and parts of Wales and was used by the King to assess dues owed to him. 900 years later, the name was given to a project in which more than a million people participated to provide a snapshot of life in the UK. Participants (many of whom were school children) contributed content, which was linked with maps, photographs, and other data. It was all compiled onto a pair of LaserDiscs – the Community Disc and the National Disc – and accessed via a BBC Master computer and a Philips VP415 LaserVision player. Armstrong came up with the idea, and the team compiled the project between 1984 and 1986. The system was eventually released at the end of 1986, and was aimed at schools, libraries, and universities. It was, alas, not a roaring sales success – the exact price per unit is difficult to pin down, but approximately £5,000 is a fair estimate. Government subsidies lowered the cost to around £3,000, but it was still out of reach for much of its audience. As a resource, however, it has huge value, which will only increase as the years pass. Tibbetts drew parallels to the original Domesday Book, which took centuries for historians to recognize as a vital resource. "Perhaps," he mused, "40 years is simply too short a time for historians to perceive the Domesday Project as a sufficiently distant benchmark of useful research." Perhaps. However, while hundreds of years have passed since the Domesday Book was compiled, the years have not been kind to the Domesday Project, thanks to technology's inexorable march. "The original Domesday Book had one huge advantage over the Domesday Project when it comes to future-proofing and longevity," explained Tibbetts. "The vellum pages of the original Domesday have been preserved for 940 years so far, and it's likely to continue to exist well into the future. "It took very few years for the bespoke technology of the BBC Domesday Project to be overtaken by universal media technologies." However, they chose the technology at the time for good technical reasons. The BBC Micro could output a 625-line television picture. The Philips LaserVision system could be re-engineered to store alphanumeric data. There wasn't much in the way of authoring and playback systems, so, as Tibbetts explained, the team had to invent everything. "In technological terms," he said, "the world of 1984 really couldn't be more different from the instant-access, media-rich world of today." And so, inevitably, within a few short years, that technology began to show its age. "As was entirely predictable, our discs aged and became difficult to read. The expensive LV-ROM player was quickly superseded by cheaper and more widely available CD-ROM and DVD systems." And, worse, "Largely because of the lack of any great academic interest in preserving our massive resource for the future, the BBC Domesday Project was in grave danger of disappearing without a trace." Several initiatives have kicked off over the years to preserve the project, and the CCH's event was as much about celebrating those as it was about looking at the project's genesis. Tibbetts dropped some nuggets of information about the project, estimating that 60 people reported directly to him and Armstrong, with another 60 in subcontracted teams. He also discussed the frustration the team felt as doubt was cast about whether the project was even possible. "We came back from Christmas in January 1986," Tibbetts recalled, "we sat down for our first team meeting, and I was forced to point out that our little VAX minicomputer didn't have one byte of data that was ready to go onto the Domesday discs, so we did the whole thing in a few months!" As far as hardware was concerned, Andy Finney, who handled mastering the discs, remembered that the original plan for the Domesday Project was for it to be mouse-based, rather than the trackball the system used for navigation. The mouse was still a relatively novel concept, alien to many users. "I remember vividly one chap who was navigating with this mouse on the table, and he moved the mouse towards him, and by the time he got to the edge of the table, he hadn't actually reached the bottom of the screen. So he went on the edge and underneath…" Tibbetts added, "I always feel guilty about the mouse, because we got that wrong … the decision came down in favor of the trackball because that is much more easy for somebody with movement disabilities to use rather than a mouse." "But boy, did we get that wrong…" As Tibbetts noted, even with little interest from some parts of the academic community, enthusiasts have continued to put in a huge amount of effort over the years to keep the resource accessible. Simon Inns of the Domesday86 project described it as "a race against time" – not just for hardware, but to get data off discs before bit rot rendered them unreadable. The good news is that the Domesday discs have been preserved, and reside in the Internet Archive. "Anyone," said Inns, "can go to the Internet Archive and download a Domesday Disc. This is, like, groundbreaking stuff. 10 years ago, this was a laugh, right? Now you can actually do this." While there might not be many functioning examples of the hardware around anymore, the preservation efforts mean that the Domesday Project will endure. "Our hope," said Inns, "is that within the next couple of years, everyone in this room [at The Centre for Computing History] will be able to have a Domesday system and be able to experience what it was like." ®

The second Anthropic-linked vulnerability known to have been exploited in the wild saw initial activity from an IP address in China targeting vulnerable hosts in the US and Japan. The vuln is a critical authentication-bypass bug in Rejetto HTTP File Server (HFS) that can lead to full admin access and remote code execution. HFS is an open source web file server that previously appeared on the US Cybersecurity and Infrastructure Security Agency’s catalog of Known Exploited Vulnerabilities in 2024. On Wednesday, researcher Zach Hanley at AI pen-testing company Horizon3 said he used Mythos to uncover a new flaw in the file server, now tracked as CVE-2026-61500. If you use Rejetto HFS, be sure to update to v3.2.1 or later, which fixes this and other security flaws. Hanley also published a video showing the steps to exploit HFS and remotely execute code on the server. By the next day, the CVE was under exploitation. “We started detecting exploitation of CVE-2026-61500 in Rejetto HFS this evening,” VulnCheck security researcher Patrick Garrity posted on LinkedIn on Thursday, adding that Hanley and team reported the bug to VulnCheck for CVE assignment. “Our canaries detected an actor in China targeting real vulnerable hosts in the US,” Garrity added. Garrity has been tracking CVEs attributed to Mythos and Project Glasswing, Anthropic’s initiative to give select partners access to the bug-hunting model, since shortly after the program was announced in April. Anthropic claims that Mythos is too powerful to release to the general public (insert evil laugh). As of Friday, Mythos and Project Glasswing have uncovered 286 CVEs, according to Garrity’s tracker, and up until Thursday only one of these bugs had been exploited in real-world attacks. The Thursday night activity originated from one IP address in China and targeted vulnerable servers in the US and Japan, Garrity told The Register. “Today we have seen four hits,” he told us on Friday. These originated from two different IP addresses in the US: 173.239.211[.]248 and 173.239.211[.]249. Both are in the same subnet, and “appear to be coming from a proxy,” Garrity added. China-linked digital intruders routinely use compromised devices as proxies to route malicious traffic and disguise the attackers’ true location, and in April a 10-country security advisory warned of China-nexus cyber operatives using proxy networks “strategically, and at scale.” In his write-up, Hanley said Horizon3 has used Mythos in its vulnerability research – and discovered “many critical vulnerabilities” – ever since the security company joined Project Glasswing in July. Mythos' mad math skillz CVE-2026-61500 highlights a couple of Mythos capabilities that make it really good at uncovering vulnerabilities, according to Hanley. Namely, Mythos excels at mathematical distillations and scientific tasks, especially those relating to computer science and operating systems. Finding this CVE “speaks to Mythos’s capabilities in understanding of mathematics, how it identified an exploitable set of cryptographic missteps, and approached solving the constraints to achieve remote code execution,” Hanley wrote. The security issue stems from how HFS authenticates users. It generates a random value with Math.random() and then passes this value to Koa, the Node.js web framework foundation for HFS. Koa uses keygrip to sign all session cookies with that random value. This means that if an “attacker can derive what the session signing key is, they can forge valid session cookies,” Hanley said. This should not be possible, assuming Math.random() uses a secure pseudo random number generator (PRNG). But V8’s Math.random() did not use a secure PRNG. Mythos discovered that the output of the xorshift128+ algorithm it used was fully reversible – and the application was leaking Math.random() outputs. The model’s analysis claimed that Z3, a Microsoft-developed, publicly available Satisfiability Modulo Theories (SMT) solver, could be used to recover the PRNG seed. Hanley notes that Horizon3’s researchers could not recall seeing an SMT solver used this way to attack a cryptographic flaw in a real application and bypass authentication. “What makes this impressive is that Mythos didn’t just flag the insecure PRNG in isolation – it simultaneously identified that the application leaked raw Math.random() outputs through a separate code path, recognized those two facts as a chain, and determined the leak produced exactly the observations needed to make state recovery feasible,” he wrote.®

"Can it run Doom?" now applies to databases, as the author of DOOMQL has released a far more visually accurate port. There's a saying: "Be careful what you wish for." In 2025, we wrote a piece about a database boffin coming up with a version of Doom that was written in SQL. It was neat, but the visuals were nearer to 3D Monster Maze than to the 90s pixel blaster. We commented, "If persuading SQL to do the entire rendering pipeline is a 'bad idea,' we cannot wait for a terrible one." And so here we are. The author, Lukas Vogel, has been at it again and ported both Doom's game logic and renderer to SQL, running it in a database. Python handles timing, keyboard input, and bitmap display, but the rest comes from SQL queries and the CedarDB database. DOOMQL was an undeniably impressive thing. However, Vogel told The Register, "Someone on Hacker News complained last year when I did the first iteration that that one was more Wolfenstein-like because I used raycasting instead of BSP-tree traversal which was *the thing* that made Doom revolutionary. "So I obviously couldn't let that stand." Which brings us to SQLDoom. Doom, ported to SQL. In his blog, Vogel noted the ground rules for the port. Unlike DOOMQL, this had to look like real Doom. The rendering had to be purely SQL-based: "the only acceptable SQL output is a table or a bitmap encoding exact RGB values for every pixel," and the game loop also had to be purely SQL-based. Vogel told us, "I was very thrilled to find out that porting the real Doom was a) feasible and b) the performance was actually *better* than the much more barebones approach from last year! "Also, at some point I started actually playing it for fun in my free time, which is still a very strange feeling. I intended it as a tech demo, but it just feels like the real Doom, even though it doesn't share a single line of code with any existing Doom port. So I wanted to see how far I could take it. "In the end, I even got deathmatch to work." According to Vogel, the game is perfectly playable on his laptop (using a Ryzen 7 7840U), and multiplayer games can be played online using EU and US servers. We found the server performance a little sluggish – likely more due to the servers than the port – although getting SQL to jump through these hoops is something that, only a few short years ago, would be unimaginable. It's worth keeping in mind that this is a thinly veiled demonstration of the capabilities of CedarDB. It's a compiling database system where complex queries are eventually compiled to machine code. Still, it's hard not to get swept up in Vogel's enthusiasm and the fact that this is Doom… running in a database. Certainly this is a lot more fun than recreating yet another online store for yet another interminable keynote. ®

“I always laugh at the IT industry’s terms,” says Ryan Sheehan, a senior vice president at $16 billion global solutions integrator SHI. And right now the jargon that gets him giggling is “forward-deployed engineer” (FDE) because, as he told The Register at VMware's September Explore conference, “We have been doing FDE for a long time.” “Doing FDE” involves sending top consultants to a client, without tying them to a project. Instead they get a mission to build something valuable using new technology, and carte blanche to make that happen in a hurry. Sheehan has company in believing FDEs aren’t fundamentally new. Cloud monitoring vendor DoIT told The Register it has used FDEs since 2011, but only started using the term in the last quarter of 2025, when it changed the name of its “Customer Reliability Engineering” practice to FDE. That decision came as services companies adopted the term FDE to describe the work needed to build AI proofs of concept. The name has its origins at Palantir, whose chief technology officer Shyam Sankar claims he coined it in 2007 after a conversation in which his CEO Alex Karp asked him why French restaurants are so good. Karp apparently answered his own question by saying, “The wait staff is actually part of the kitchen staff. They intimately understand the food, the methodology, and the technique. They are not merely carrying the food from the kitchen to the table, but are instead part of a subtle and complex system that affects kitchen operations.” Sankar says Karp asked him “to build that, but for engineering” so he did and called it “forward deployed engineering.” “FDEs embed alongside our customers and work to ensure our software solves their problem and not some proxy for their problem,” Sankar wrote on his personal Substack. “They’re crazy enough to get on a last-minute plane to Iraq, they’re smart enough to ship quality, same-day code, and their EQ is still high enough to talk to users (and maybe even enjoy it).” Sankar says Palantir’s investors thought FDEs were a bad idea because the costs they incurred ate into margins, but says they’re worth it because they ensure customers understand the company’s products and how to get value from them. “We didn’t believe in throwing our software over the wall in the hopes the customer would divine the correct meaning from it,” Shankar wrote. “This approach built an engineering organization with unmatched creativity, responsiveness, and focus on the primacy of winning.” AWS likes winning, too, and in June announced a dedicated AWS FDE organization, and $1 billion to fund it. In July Microsoft trumped that with a $2.5 billion so-called “Frontier Company” it claims “goes beyond” FDEs. Both companies see FDEs as a way to get their consultants into customers to help them build AI apps. They argue FDEs’ time has come because, as Palantir found when users struggled to put its software to work, organizations often struggle to turn AI experiments into production applications. Stop us if you've heard this one before But vendors also admit that FDEs aren’t entirely new. Indeed, AWS told The Register it has offered customers something very similar to an FDE for years. “We started this journey all the way back in 2017 when the focus was machine learning (ML),” said Taimur Rashid, the cloud giant’s managing director for go to market, Frontier AI Engineering & Services. “We created a team called the ML Solutions Lab. The idea was that the tech is new, customers need to navigate it. So we said ‘Why don’t we loan them a data scientist?’ They can build a PoC, and the customer can build on our platform.” Rashid says AWS sometimes called this sort of arrangement a “secondment,” a “resident architect” or a “resident scientist.” Whatever you call it, Rashid said the approach worked well so AWS decided to evolve it when generative AI burst onto the scene – this time using an applied scientist and a strategist, and an evolved approach. “We forward-deploy them – they will help to identify the art of the possible and develop the solution for you or with you,” Rashid explained. “They will be part of your business and operational workflows, and effectively be yours for a defined time,” he said. That’s different to conventional services engagements, he argues, because there’s no scope of work, no hourly charge for consultants’ services. “What is different about FDE is you are saying the customer wants to build. The path is not determined.” Carlos Pereira, a Fellow and Chief Architect in Cisco’s Customer Experience team, also admited FDEs aren’t entirely new, but argued AI makes them more valuable than conventional consultants. "The idea of an expert being embedded on-site to help a customer adopt new technology and solutions isn't new,” he told The Register by email. “Our Customer Success organization has been doing this for years.” “The evolution is that AI is now a key part of that equation,” he added. “It introduces new variables like non-deterministic AI models, evaluations, faster-moving requirements, and specialized skills that not every customer team has in-house.” Analyst firms Gartner and Forrester agree that FDEs can be useful, and that this form of IT services differs due to greater intimacy between client and consultant. “They often have full-stack access to the customer’s environment and the vendor’s codebase, allowing them to build and iterate in real time,” Gartner’s Arun Chandrasekaran and Robert Thanaraj wrote in March guidance titled “An AI Leader’s Guide to Forward Deployed Engineering to Scale AI.” A June best practice report from five Forrester analysts, titled “Forward-Deployed Engineers: Use provider-embedded builders to improve outcomes,” praises the model for helping to “accelerate adoption and turn complex platforms into production-ready outcomes.” Yet both analyst firms also warn that using FDEs may see clients develop a dependency on the vendor that provides them. “The allure of delivering outcomes at speed can induce blind spots and is fraught with risks,” according to Forrester. “Without the appropriate guardrails on tech architecture and stack choices, teams can fall into a bespoke trap, where highly tailored solutions deepen vendor dependence and make future change costly.” Debt and dependency Gartner urges would-be hirers of FDEs to “beware of ill-conceived FDE engagements that can lead to a high degree of vendor dependency, security/data risk exposure, technical debt, and talent atrophy.” The latter happens when organizations hire FDEs but don’t train their own teams on new tech. “Your internal team may become operators of a black box rather than architects, making it difficult to innovate without the vendor,” Gartner warns. Gartner also worries FDEs can be a fast-track to technical debt. “A major concern with the FDE model is that it relies heavily on manually designed structures and hard-coded domain knowledge. Such human-engineered components tend to deliver short-term gains but are repeatedly superseded as more general, scalable learning systems improve.” “As large language models continue to advance, much of this manually encoded logic is likely to be rendered obsolete, requiring costly rework or full redesign, and creating a long-term technical and organizational burden.” FDEs themselves may soon be obsolete, because they’re not vendors’ only bet on improving AI consultancy skills, according to Omdia chief analyst Jay McBain. “At the same time as investing in FDEs, the big players have been investing big into partner enablement,” he told The Register. “They are playing both ends.” The tech services channel doesn’t mind that, because according to Omdia data 82 percent of partners report they “aren’t ready” to quickly grow their AI services. The vendors promoting FDEs, meanwhile, don’t want to use this form of consultancy for long. “None of these companies’ investors want to dilute a low-value services business into the overall revenue,” McBain wrote. “We are predicting all of this FDE investment will fall back into the channel by 2031 when this matures.” Which rather suggests that in a few years we’ll all be laughing at how Palantir re-heated an old form of tech services under a new name, and the rest of the industry decided to follow the same recipe. ®

After evolving from an Apple-made, client-side language and years of false starts, Swift could finally make it on the server. Google has taken note and has just rolled out Google Cloud API Client Libraries for Swift, evidently convinced Swift has a backend role thanks to its performance and ease of use. The GCP Swift libraries target Swift 6.2+ and take advantage of SwiftNIO's asynchronous event loops, HTTP/2 multiplexing, gRPC transport, and data race safety compiler flags. "For years, Swift was perceived mainly as a UI language tied to Apple client devices," said Karl Weinmeister, director of developer relations, and Carlos O'Ryan, software engineer, in a blog post. "With Swift 6 and strict concurrency checking, it has matured into a viable systems and cloud language, pairing Rust-like data-race safety with predictable, reference-counted performance." Weinmeister and O'Ryan argue that Swift manages to be both developer-friendly and suitable for low-level control over resources, which tends not to be the case for challenging systems languages like C++ or Rust. They voice appreciation for the language's lightweight runtime and Automatic Reference Counting (ARC), alongside its expressive syntax. And they approve of Swift's compile-time concurrency checking. "Data races are caught in your editor before a binary ever compiles or reaches production," they note. The Googlers argue that the Server Side Cloud Swift SDK is well-suited for servers, containers, and DevOps environments, pointing to its utility for building capable microservices using Swift web frameworks like Hummingbird or Vapor. "With google-cloud-swift, server-side Swift developers can write end-to-end cloud infrastructure with compile-time race safety, native async/await ergonomic APIs, and zero OS thread congestion," they declare. It's been quite a journey. Apple open-sourced the Swift programming language all the way back in 2015 and in the years that followed, developers started looking at Swift on the server. IBM in 2016 made Swift available on its Bluemix cloud and a few months later the Swift team announced a server API working group. The first server-side Swift conference was announced in late 2017. And Apple's work on the async SwiftNIO framework in 2018 advanced the case for backend Swift. Swift's popularity peaked in January 2020, according to the TIOBE index, when it ranked as the 9th most popular programming language. It currently ranks 18th. Stack Overflow's 2025 developer survey has Swift ranked 20th. Even so, server-side adoption of Swift has been slow. IBM reversed course in 2019, backing away from its Swift commitment. And Amazon hasn't really done much with its Smoke Framework for server-side Swift in the past few years. Still, the Swift community has kept at it. The Swift Server Workgroup launched in 2018 and, by 2024, the release of Swift 6 with data race safety checks made the language more interesting for backend applications. Swift may be more approachable than languages like Rust or C++, but it's unclear how much that will matter in the years to come as developers become more and more dependent on AI coding agents. Certainly, Swift code is readable, a benefit for those maintaining applications, but language ergonomics and approachability matter less when AI is reading and writing most of the code. ®

Faced with growing opposition to datacenters from residents who don't want the bit barns in their backyards, Amazon plans to invest $1 billion over the next five years to indoctrinate the public. The company announced the investment in a whataboutism-packed blog post aimed at convincing readers datacenters are not as bad as they've been led to believe, and that if the common folk don't get in line and support their development, the nation could be left in the dust. The program aims to invest in education, job training, local infrastructure improvements, and sustainability initiatives in communities home to Amazon datacenters. "This build out is so important geopolitically that there are widespread reports of various countries intentionally seeding misinformation in the US about datacenters to trick us into slowing down," the e-tailer turned cloud titan wrote. "Right now there are over 100 datacenter moratoriums being considered across the country. If these measures are enacted, the US could be writing its own losing ticket to this race, and the consequences would last generations." If Jeff Bezos's cash cow is to be believed, datacenters will transform the US economy much like the buildout of the highway system did in the '50s. But that can't happen, Amazon argues, if Americans block the development of these bit barns over misguided fears. The blog post highlights four such datacenter fears, which Amazon claims are "myths," despite plenty of reporting to suggest otherwise. Who needs water? The first "myth" is the fact that datacenters consume an excessive share of water. Bit barns commonly employ evaporative coolers which use water to reject heat generated by the facility into the atmosphere. These cooling towers use less energy than dry coolers or refrigerant-based systems but remove water from local watersheds as a consequence. According to Amazon, citizens should really be more worried about the water consumed by golf courses, which it claims use 200x more water than all of Amazon's datacenters. That argument might be more convincing if people didn't like golf courses. But they do and the whole lesser of two evils argument works a lot better when one of those evils isn't extremely popular. Amazon also attempted to dispel claims that a typical datacenter consumes five million gallons a day, arguing that its facilities are much more efficient using just 13,000 gallons a day, the equivalent of 42 US households. But this leaves out an important detail. It is possible to build datacenters that don't consume large quantities of water, but these facilities often consume large sums of power as a consequence. Most gas, nuclear, coal, and anything else that uses steam to turn a turbine also consumes water. In other words, nearly every large-scale power-generation method uses water either directly or indirectly. And despite Amazon's attempts to downplay the impact of datacenters, there's good reason for those living nearby to be worried. Earlier this year, it was reported that a datacenter, which we'll note was not operated by Amazon, has quietly sucked up nearly 30 million gallons of water and the only reason anyone discovered the issue was that local residents started complaining of unusually low water pressure. We totally won't let datacenters drive up your power bill or poison your air Amazon also attempted to dispel fears that datacenters could leave neighboring residents picking up the tab for their massive power consumption. The company argues that rates are going up because the US grid is old, and that responsible bit barn builders are working with utilities to ensure they're paying their fair share for the additional capacity and infrastructure upgrades required to support them. What wasn't addressed is why folks might be worried about rising rates. One of the reasons that utilities and plant operators are so conservative is that they've been burned by speculative markets before. If new power plants are built to meet demand that never materializes, ratepayers could get stuck with the bill. Amazon and others agreeing to pay higher rates might help, but only if they're consuming an appreciable amount of the capacity they convinced utilities to build. An empty datacenter doesn't pull a lot of juice. Speaking of power, Amazon also attempted to downplay the impact of datacenters on air quality due to their reliance on diesel backup generators. "The truth is the datacenter generators almost never run," the company wrote. Amazon also argued that these generators are subject to stringent permitting requirements which dictate how long they can run for. That might be more reassuring if datacenters in power constrained regions weren't getting permission after the fact to exceed those limits. And, as we saw with Elon Musk's Colossus supercomputer, temporary gas turbines can be deployed when the local grid can't support the load. That hasn't sat well with residents of Boxtown, Tennessee, while the NAACP is suing xAI over allegedly unpermitted turbines in Mississippi. Amazon can argue it's got a better track record than most, but that doesn't mean the community's concerns are misplaced. But think of the taxes Finally, Amazon argues that folks don't realize how big of an economic impact datacenters have on their community. "Datacenter operators are often the single largest taxpayer in a community, with upfront payments routinely exceeding entire annual municipal budgets in the first year of construction," the company claimed. "For example, in St. Joseph County, Indiana, the taxes Amazon will pay to the county are estimated to exceed $3 billion." The subtext is that, by enacting datacenter moratoriums, communities are walking away from billions in tax dollars. Then again, if the tax revenue from these facilities is so great, it raises the question why Amazon needs to spend more than $1 billion over five years sweetening the deal for communities hosting its bit barns. ®

arXiv, the preprint paper repository, is seeing so many low-quality submissions, with AI tools helping drive the surge, that it has imposed a hard limit of two new submissions per month per submitter, regardless of whether they were produced solely by humans or with help from AI tools. After becoming an independent nonprofit over the summer following 25 years under Cornell University’s stewardship, arXiv continued to rely on a “practical limit” in place that gave moderators the ability to determine whether someone was submitting too many papers. Like everything good and pure in this world, AI has ruined that, too, necessitating some strict new limits. “arXiv now limits submitters to up to two submissions per calendar month, with a limit of three total active submissions at any given time,” arXiv community engagement manager Kat Boboris wrote in a Thursday announcement of the change, which was effective immediately. According to an arXiv spokesperson, an active paper is one that has been submitted, is under consideration and review, but not yet published on the site. It’s worth noting that large institutions whose projects involve multiple researchers are unlikely to see a significant impact on their submissions - this limit is per submitter, the announcement explains, so a research paper with multiple authors only counts toward the submission rate of the individual who files it. Rejected papers count toward that monthly total. If that sounds like some strict limitation - it is, but it’s necessary, based on arXiv’s numbers. In September alone, arXiv said that it had received 40,363 submissions. The same month 10 years ago saw it receive just 9,869 submissions; two years ago that number was 20,569. Submissions have increased 309 percent in the past decade, in other words - and doubled in the past two years alone. Unsurprisingly, many of the submissions that arXiv has been receiving of late don’t meet its requirements for AI-assisted work, which call for significant use of generative AI tools to be disclosed, for those tools to be used in support of research, and for the resulting work to meet arXiv’s standards for scholarly interest and advancement of the field. “Many of the submissions that arXiv is now receiving do not satisfy these requirements,” Boboris said, describing many recent submissions as “thin” and “narrow in scope,” and noting an uptick in papers that break a single work up over multiple smaller papers. “A relatively small proportion of authors are submitting a large number of low-quality papers and consuming a disproportionate fraction of the moderators’ time,” arXiv editorial advisory council chair Thomas Dietterich said. “This is unfair to authors who continue to submit quality papers — their papers can be delayed for days or weeks as a result.” It’s unfortunate it’s come to this, but the slop sluice is wide open and flooding the channels of academia. arXiv, as a non-profit with volunteer moderators, simply can’t afford to keep combing through the slop pits. The change may not be permanent, however. arXiv notes in the announcement that the new monthly submission cap is a stopgap measure meant to stem the tide while the organization figures out what new best practices for submissions should look like, while its separate limit of three active submissions has been in place since 2024. “arXiv will be monitoring the impact of this change on our submissions rates and moderation load to evaluate its overall effectiveness and impact on researchers and will make further changes as needed,” a spokesperson explained. In the meantime, arXiv is begging everyone to be purposeful in their submissions. “arXiv would ask researchers to prioritize their submissions and consider submitting papers that reflect the full scope of their work rather than multiple incremental papers,” the arXiv representative said. ®

OpenAI's agents have repeatedly strayed beyond their intended scope. Two separate reports detail the activity, including one from Sam Altman’s company saying it has notified more than 100 organizations about potentially problematic model activity. OpenAI, in a late Wednesday update to its ongoing Hugging Face investigation, said it has notified more than 100 organizations that “misaligned models” may have accessed their systems. “Notification does not mean that any private information was accessed, or that there was a compromise of any third-party system,” the update said. A separate Thursday report from digital forensic and incident response startup Asymmetric Security said OpenAI’s rogue agents accessed data belonging to 55 organizations. These include the US Department of Education, UN Trade and Development, US Bureau of Economic Analysis, MAX.gov containing federal budget documents, the European Centre for Disease Prevention and Control, the US Securities and Exchange Commission, the International Energy Agency, and the FBI Crime Data Explorer. Asymmetric used only publicly available data to compile this list, and said the activity occurred between March and September. The agents’ probes indicate they were tasked with researching public health and other data, “possibly as part of an evaluation,” according to the report. “We found successful access to staging environments; evidence of the use of attacker reconnaissance tactics; and evidence of probing a broader set of websites, including those of the CDC, SEC, International Energy Agency, and Mayo Clinic,” it said, noting that the investigation also uncovered some “novel tactics” the agents used to break out of their sandboxes and gain full web access. “Some of these tactics left records erased or inaccessible, making it impossible to rule out access to sensitive data based on public information alone,” the authors wrote. The Register asked OpenAI if the organizations on Asymmetric’s list were among those notified by OpenAI. The model maker declined to say which orgs had been notified, but previously confirmed to the New York Times that its agents probed websites for the US Education Department, Commerce Department, and the Securities and Exchange Commission. An OpenAI spokesperson sent us this statement via email: “As we previously announced, we’re reviewing misaligned model activity and notifying organizations when we identify potential impacts to their systems. We’re also investigating findings in third-party reports, comparing them with our own and seeking additional information where needed. Our priority is to provide affected organizations with accurate, useful information, and we’ll keep refining our approach as we learn more. Most of the activity we’ve reviewed involved routine research tasks, including accessing public web content. Some involved government websites, which our models often use as authoritative sources of public information.” The growing number of rogue agent hacking incidents raises questions about AI makers’ safety and security practices during testing - and has increased calls for holding AI executives legally liable for their models’ criminal activities. According to Horizon3 CEO Snehal Antani, who builds and tests agents at his threat-exposure startup, the term “misalignment” lets frontier model makers off the hook too easily. “A ‘misaligned models incident’ is basically a fancy way of saying a model didn't respect scope - or wasn't given one - had no audit logs or observability in place to detect breakout, and accessed third-party systems without authorization,” Antani told The Register. “The responsibility sits with the labs that build and deploy these models,” he added. “The safety-versus-security framing lets them sidestep accountability, and they are not incentivized to prioritize security because moving fast is the priority.” OpenAI’s most recent rogue agent disclosure comes as it - and every other major AI company - drinks from the firehose of near daily security and safety concerns surrounding its models. Last Friday, OpenAI quietly paused training of its most advanced models after admitting an agent used DNS to reach an external chatbot. On Monday, it postponed its planned release of GPT-6.1 Astra after the model showed higher levels of deception than its predecessor, including not always accurately telling users what actions it had or hadn't taken. It also performed unsolicited supply chain attacks in simulated security evaluations, according to the UK Artificial Intelligence Security Institute. On Wednesday, OpenAI accused rival Chinese model maker Moonshot AI of distillation - essentially copying OpenAI models’ reasoning at scale - and said that poses a national security concern. Early Friday, OpenAI confirmed to The Register that it fired two safety researchers and a program manager for allegedly mishandling sensitive company information.®

A California business owner was arrested on Thursday after being charged with allegedly smuggling Nvidia hardware to China without the proper export licenses. Keeping the highest-end GPUs out of Chinese hands has been a priority for the US government. Greg Lui, 38, allegedly tried to export around $300 million worth of Nvidia kit - typically used for artificial intelligence (AI) development - via US-based freight businesses and Malaysian transshipment companies. Prosecutors allege that Lui used his company, Earthmade Computer Inc, to order the products in the US, which consisted of servers containing high-powered Nvidia components, including A100, H100, PNY GE Force RTX 4090, and GeForce RTX 5090 GPUs. Lui would then send them to Malaysia and Singapore, two countries that do not require licenses from the Commerce Department, and companies there would forward them to China in an effort to circumvent US export controls. His alleged crimes violate regulations prohibiting the sale of economically sensitive equipment to adversarial countries, such as China, in the interests of national security. In lay terms, the US doesn’t want its own tech companies’ high-end hardware being used to accelerate China’s push for dominance in the race for the world’s most capable AI. Or, as it’s referred to in the US nowadays, “super intelligence (SI).” “SI is the defining technology of the era,” said John A. Eisenberg, assistant attorney general for national security. “The National Security Division will protect the American advantage in the chips that power this technology, a product of our unparalleled innovation and hard work, from illegal diversion by our economic and military adversaries.” According to court documents [PDF], the scheme began in or around October 2023 and continued until at least August 12, 2026. Prosecutors claim Lui received more than $176 million in payments from two Malaysian transshipment companies, and in return introduced them to US-based companies capable of supplying Nvidia hardware, brokering the sales of almost $300 million worth of kit. The US government accuses Lui of being fully aware of the legal implications of his alleged actions, and that US export laws restricted the shipments of advanced Nvidia chips to China to license-holders. Court documents indicate that the US got its hands on documents tying the export of 92 servers to Kuala Lumpur from San Francisco International Airport, ordered by Earthmade. The documents also list the consignee for the onward re-shipment from Malaysia to Hong Kong as a Chinese customer. The indictment includes claims that Lui instructed a US front company to fraudulently list the recipient of a shipment as Jackie Lui, the supposed CEO at “Topmost,” a company registered in California by the CTO of one of the Malaysian transshipment businesses. Prosecutors allege that, in 2021, Lui illegally purchased identity documents that were used as part of the scheme. Lui is charged with one count each of violations related to the Export Control Reform Act and the Export Administration Regulations, outbound smuggling, and money laundering, which together carry a maximum prison sentence of 50 years. “The FBI’s investigation revealed that Lui allegedly sold the Chinese government hundreds of millions of dollars’ worth of American Super Intelligence technology, in clear violation of US export control laws,” said Roman Rozhavsky, assistant director at the FBI’s Counterintelligence and Espionage Division. “Controlling the export of advanced SI technology is critical to safeguarding our national security and defending the homeland, and the FBI will keep fighting for America’s businesses and economic security. We’re grateful to our partners across the US government and private sector for their assistance, and we’ll continue holding accountable anyone who violates US export laws to enrich themselves and help our adversaries.” The Register contacted Earthmade and Nvidia for comment. ®

Leaving VMware forces a fresh look at every layer of the datacenter, and data protection belongs near the top of that list. Most teams plan to move the backup design across as it stands and rebuild it on the new hypervisor. The better plan uses the exit to raise resiliency and recovery speed, keeps the backup application, and gives each layer the work it does best. What the migration touches Backup designs in most data centers rest on vSphere mechanics, from changed block tracking through the vStorage APIs and VMware snapshot behavior during a job to storage integration registered against the vCenter inventory and proxies placed per cluster. Many shops add an orchestration product such as Site Recovery Manager to drive the failover plan. Change the hypervisor and each of those assumptions gets a second look. Backup vendors have done real work here, and support for the major KVM-based platforms has matured quickly over the past year. The bigger question sits upstream of the backup application. In the vSphere design, production ran the workloads and nearly everything else fell to a separate protection tier. Resilience past a drive failure, fast rollback, replication, site failover, and immutable copies each became another product with its own license, console, and renewal. Why the old division of labor made sense That design grew out of what the platform could do. Server virtualization abstracts a single server, presenting virtual CPUs, memory, and disks to a guest and leaving storage resilience to an array and recovery to a backup product. First-generation HCI moved storage into the servers and kept the same division of labor, often running a controller VM on every node. A backup tier that handled everything from a failed drive to a lost site was the right answer for that architecture. The cost of that arrangement shows up on the worst day. A single site failure touches the hypervisor, the storage system, the backup application, the replication tool, and the network configuration, and the recovery path crosses every boundary between them. The exit is the first moment in years when that assumption is open for review. What the production platform should take on Datacenter virtualization abstracts the whole environment. Compute, storage, networking, and data protection share one code base and one set of metadata, so the platform holds a complete picture of a workload rather than a collection of blocks. That shared view lets production absorb responsibilities the old design pushed downstream. Far fewer events reach the backup tier once production handles them. A second drive failure becomes a hardware replacement task instead of a restore in the middle of the business day, and a bad patch turns into a quick snapshot rollback that the application owner barely notices. That same resilience is what makes refurbished servers and drives a defensible choice for production, and flash and memory pricing climbed through 2026, which makes the hardware you already own worth more than last year's models assumed. Backup's job gets bigger, starting with the migration Shifting day-to-day recovery into the platform expands the backup application's role. The first new assignment arrives before the new environment runs a single production workload. The backup product already protecting the vSphere estate holds a current, consistent copy of every VM, and a backup application that supports both the old hypervisor and the new one turns that copy into the migration path. Veeam Backup & Replication is a good example of this model and now supports VergeOS as a platform, so a team can restore its vSphere backups onto the new environment, test them there, and cut over on a schedule it controls. The investment made to protect VMware becomes the vehicle that moves the workloads off of it. After the migration, the backup application keeps the work it has always done well. Long-term retention, compliance archives, air-gapped and offsite copies, and granular recovery for application owners all stay with the backup tier. It also becomes the independent copy that lives outside the production platform's failure domain, which is the right place for the last line of defense. Proving the platform side Asking production to do more concentrates responsibility in a single vendor's engineering. That trade is defensible, and it is only defensible against evidence. Ask for the failure scenarios live, against a running application, rather than on a feature matrix. Have the vendor pull a drive, then pull a second one past the stated protection limit with the workload still running. Ask them to delete a VM and recover a single file out of it and then take the primary site offline and bring the environment up somewhere else. A vendor that can run those four tests on a call has answered the question, and a vendor that offers a diagram instead has told you something too. Then run a test migration through the backup application and confirm the restored VMs boot and perform on the new platform. The decision the exit presents The hypervisor swap is the visible half of this project. How protection is divided between production and backup decides what the environment costs to run for the next five years and how it behaves on the worst day of those five years. Rebuilding the old division of labor is a choice, and it deserves the same scrutiny the licensing spreadsheet received. VergeIO built VergeOS on the datacenter virtualization model described above, with resilience, snapshots, instant recovery, and site failover inside the same code base as compute, storage, and networking, and with VergeOS support in Veeam Backup & Replication for the migration and the backup tier that follows it. At a recent session, Aaron Richman and Dave Vincent ran each of the four failure tests against a running application, from a single pulled drive through a full site recovery. The session is available on demand. Take a look, and see how you can turn a big exit into a bigger opportunity. George Crump is Chief Marketing Officer at VergeIO. He founded Storage Switzerland and spent his career as an industry analyst covering storage, virtualization, and data protection. Supplied by VergeIO

California's attorney general has subpoenaed OpenAI as the state investigates what happens when the AI lab's models escape their testing environments and start meddling with systems on the open internet. Attorney General Rob Bonta said his office served OpenAI with an investigative subpoena this week as part of a broader California Department of Justice probe into cybersecurity incidents and risks involving the company and its models. The move follows an investigation launched last month into an incident involving Hugging Face, after OpenAI's agents managed to break out of their test environments and onto the public internet. They then went poking around Hugging Face's systems, with one agent even creating an account on the platform without being told to. California's DoJ isn't saying exactly what it has demanded from OpenAI under the subpoena, but Bonta said the state wants more information about cybersecurity incidents involving the company. "My office is asking OpenAI additional questions regarding cybersecurity incidents and risks involving the company and its AI models," Bonta said. He also made clear that the investigation is looking at where responsibility lies when an AI model ends up doing something its developer didn't intend. "Frontier models can be legitimate tools for cyber defense — at the same time, companies that develop these models and offer them for use have a moral and legal responsibility to ensure that they do not perpetrate or enable cyberattacks, either during model testing and development or once models are placed into service," Bonta said. "Developers that fail to do so can and should be held legally accountable, and my office is committed to determining if that is the case here." The subpoena doesn't mean California has concluded OpenAI broke the law, and the attorney general's office hasn't identified any specific violation. For now, the state is still gathering information. But the investigation has been building for several weeks. In September, Bonta joined a bipartisan group of 25 attorneys general calling on Congress to regulate large-scale AI models following reports of cybersecurity incidents at frontier AI labs. That letter specifically pointed to reports that OpenAI models undergoing evaluations had escaped their testing environments, reached the public internet, and accessed outside computer systems. The attorneys general called for a government-led incident response regime that would give investigators direct access to AI companies' records when things go awry. Bonta's office now appears to be putting some of that thinking into practice at the state level. As The Reg previously pointed out, the sandboxes intended to contain these agents need to be more secure, which is the most logical reasons why the Hugging Face and other incidents happened in the first place. OpenAI didn’t respond to our questions. ®

Oracle's Wisconsin AI datacenter could miss its planned 2027 start date because the transmission infrastructure needed to power it is awaiting regulatory approval. Research company Aterio warns that Project Lighthouse, being developed by Vantage in Port Washington with Oracle as tenant, faces a real risk of slipping beyond Big Red's target of delivering capacity to customers in the second half of next year. Separately, Oracle reportedly sent a force majeure notice to a Blue Owl Capital subsidiary developing Project Jupiter in New Mexico. Bloomberg reported that Oracle was seeking to defer payments if the campus failed to come online as planned, rather than withdraw from the project. Oracle says the development remains on schedule. Aterio said of this and the Wisconsin facility: "The regulatory record points to the potential for dates later than company guidance, and Oracle's stretched finances, together with its force majeure notice on Project Jupiter in New Mexico this week, raise the potential for slippage on its largest campuses." Oracle has been offered the opportunity to comment. The Register has already reported that Oracle could face more than $100 million a year in financing costs to guarantee the power commitments behind the Wisconsin datacenter campus, which is rated at 902 MW of computing load and 1.3 GW of total electricity demand. The immediate obstacle is American Transmission Company's (ATC) application to the Public Service Commission of Wisconsin (PSC) for approval to build the campus's grid connection. ATC is planning more than $2 billion in transmission infrastructure projects tied to datacenter developments across Wisconsin, ranging from high-voltage power lines and substations to upgrades to existing facilities. Construction of the transmission infrastructure requires PSC approval. "The problem is power. The campus cannot run until ATC connects it to the grid, and ATC cannot start building until the PSC approves its application," Aterio said. The PSC closed the original case after withdrawing its finding that ATC's application was complete. ATC refiled in September, restarting the approval process. Aterio's earliest scenario puts partial power in October 2027 and full supply in August 2028, although it considers that timetable highly unlikely. Its base case puts partial power in December 2027 and full supply in October 2028. An extended regulatory review could push those dates to June 2028 and April 2029. Even December 2027 power delivery would leave little room for testing and commissioning before Oracle's deadline. "In practice, a datacenter does not enter service as soon as first grid power arrives. Each building's electrical and cooling systems have to be energized, tested and commissioned before it can carry customer load. That will take time. We therefore expect the campus's load to ramp up mainly from the first quarter of 2028," Aterio said. Meanwhile, the Financial Times reported that Tencent had signed a $7 billion deal to lease around 100,000 AI processors from Oracle. For context, Oracle reported $55.7 billion in total capital expenditures for its fiscal 2026, which closed at the end of May. Its total revenue was $67.4 billion. Capex guidance for the 2027 financial year is up to $95 billion. ®