Daily edition · 20 items · covers 3 Oct 11:50 → 4 Oct 11:15 UTC · how this edition was made
Sunday, 4 October 2026
Security 20%Military 15%Compute 15%Deployment 15%Research 10%Health 10%Policy 10%Frontier 5%
0:00 / 14:02
President Trump appointed Director of National Intelligence Jay Clayton as the administration's AI czar, to lead a new White House panel the administration calls the "Super Intelligence Force" and to report within 120 days on the risks and opportunities posed by AI, the Wall Street Journal reported on Saturday and Reuters and CNBC relayed. Treasury Secretary Scott Bessent, on The Axios Show, told AI executives who want federal guardrails to slow down on their own: "It's kind of like Hannibal Lecter: Stop me before I kill again. This alarmism without solutions by some of the AI community — that's not leadership." The White House separately published a "Kyoto Vision" endorsed by 17 countries, pledging to expand researcher access to AI-for-science tools, data, computing infrastructure and experimental facilities.
Northrop Grumman said its YFQ-48A Talon Blue "completed its first fully autonomous flight, including taxi, takeoff, in-flight maneuvers, and landing" at Mojave Air and Space Port. Microsoft and Hugging Face released ThinkingBox, which runs each of 507 stateful business tasks 20 independent times and grades on database state: Claude Opus 5.5 leads on pass@1 at 67.16%, but passes all 20 attempts on only 241 of the 507 tasks.
Google suspended product-vulnerability submissions to its open-source bug bounty on October 1 over an influx of invalid AI-driven reports. And The Register reported in-the-wild exploitation of CVE-2026-61500, a Rejetto HFS authentication bypass found using Anthropic's Mythos model — the second Anthropic-linked vulnerability known to have been exploited in the wild.
Frontier models & labs
OpenAI says it pauses training or holds back models "when we need to slow down", responding to its safety lead's resignation Company claimUpdate
- OpenAI spokesperson Drew Pusateri told TechCrunch: "We're making sure our models don't become more capable than we can safely manage and secure, and we pause training or hold back models when we need to slow down." He added that OpenAI is "making significant changes to strengthen security in our research and testing environments, train models to not just complete tasks but do so responsibly, expand our work with third-party evaluators, and improve real-time monitoring so we can detect and respond to concerning behavior earlier in the training process."
- This is the company's first on-the-record answer to David Robinson's resignation essay, covered here yesterday. TechCrunch reports Robinson said he "led the writing of safety reports that accompanied OpenAI's major product launches" and, at three-and-a-half years, is "among the longest-tenured employees at the company".
- Robinson argues frontier AI companies should run "like nuclear-power plants or busy airports, with layers of redundancy and careful, time-consuming planning, so that the occasional and inevitable human error does not open a door to disaster", and says he "never encountered a colleague who had experience making airplanes fly safely or nuclear reactors run without melting down, or helping the financial system grow without collapsing".
- OpenAI did not say which models, if any, it has paused or held back, and gave no dates or figures. The statement is the company's characterisation of its own practice and is not independently verified.
Research & papers
ThinkingBox runs each of 507 business tasks 20 times: Claude Opus 5.5 leads at 67.16% but passes all 20 on only 241 Company claimPreprint
- The Microsoft and Hugging Face post describes 507 stateful business workflows — retail 98, auto insurance 100, travel 104, neobank 104, consulting 101 — where "each task runs 20 independent times, each from an identical clean backend" and grading is on "terminal backend state and side effects" rather than on the model's transcript. Task-weighted pass@1: Claude Opus 5.5 67.16%, Claude Opus 5 66.50%, GPT-5.4 65.36%, GPT-5.6 Sol 61.91%, GPT-6 Astra 58.31%; best open-weight is Kimi-K3 at 57.37%.
- Repeating the task is what separates the models. Claude Opus 5.5 and Claude Opus 5 each pass all 20 attempts on 241 of 507 tasks (47.53%) and GPT-6 Astra on 231 (45.56%), against 128 for GPT-5.4 (25.25%) and 68 for Kimi-K3 (13.41%). The post reports that GPT-6 Astra retains 78% of its single-attempt rate across 20 repeats and Claude Opus 5.5 and Claude Opus 5 each retain 71%, while GLM-5.1, Kimi-K2.6 and DeepSeek-V4-Pro "each keep about 8%".
- Most failures look like successes from outside. In a common-set ablation of 121,680 valid trials across 12 models, 79,853 attempts failed the executable checks, and 67.24% of those failures "still terminated cleanly, invoked a state-changing tool, and reported no final tool error". The checks found wrong field values in 77.61% of failures, unintended extra effects in 43.30% and missing required effects in 25.36%, categories that overlap.
- This is a Microsoft-run evaluation of rivals' models published by Microsoft, and the underlying paper, arXiv:2608.19741, is a preprint. The cost figures in the post are described as "Estimated $, not actual cloud bills" and are priced at undiscounted OpenRouter list rates.
Exgentic publishes 10K real agent runs and 242K model calls across five frontier models in OpenTelemetry format beneficialCompany claimSingle source
- Lena Dankin, Elron Bandel and Michal Shmueli-Scheuer released "10K runs and 242K model calls, filtered down from a full corpus of 10.5K runs and 626K calls", with scaffolding calls — simulated users, automatic graders, repeated identical retries — stripped so the totals describe only the model under test.
- Coverage is DeepSeek-V3.2 (2.3K runs), Kimi-K2.5 (2.3K), GPT-5.2 (2.1K), Claude Opus 4.5 (1.9K) and Gemini 3 Pro (1.4K), across SWE-bench (2K runs, 2.1M average tokens per run), BrowseComp Plus (1.9K, 1.6M), AppWorld (1.5K, 1.7M) and three τ²-bench customer-service variants (telecom 1.8K at 332K, retail 1.8K at 198K, airline 1K at 255K).
- The spread between models is large on the same work: Claude Opus 4.5 averages 2.4M tokens per run against GPT-5.2's 552K. Every call is written in the OpenTelemetry GenAI conventions with full input and output messages, available tools, token usage, stop reason and error type.
- These are the authors' own figures for their own dataset and no independent audit is reported. It is a data release, not a paper, and carries no arXiv ID.
Security, misuse & threat intelligence
Rejetto HFS flaw found with Anthropic's Mythos model is exploited in the wild, the second such bug known to be attacked harmfulCompany claimUpdate
- The Register reports that CVE-2026-61500, a critical authentication bypass in Rejetto HTTP File Server that yields admin access and remote code execution, is now under attack — "the second Anthropic-linked vulnerability known to have been exploited in the wild". The fix is HFS v3.2.1 or later.
- VulnCheck researcher Patrick Garrity told The Register the first activity came on Thursday evening from a single IP address in China, targeting vulnerable servers in the US and Japan: "Our canaries detected an actor in China targeting real vulnerable hosts in the US." On Friday, Garrity said, "we have seen four hits", from two US addresses in the same subnet, 173.239.211[.]248 and 173.239.211[.]249, which "appear to be coming from a proxy".
- Per Garrity's tracker, as cited by The Register, Mythos and Anthropic's Project Glasswing have uncovered 286 CVEs as of Friday, and until Thursday only one of those bugs had been exploited in real-world attacks. Horizon3.ai's Zach Hanley says the server derives its session-cookie signing key from JavaScript's Math.random(), leaks further Math.random() outputs through a separate unauthenticated code path, and that Mythos chained the two to recover the generator state and forge admin cookies.
- The attribution of the discovery to Mythos, and the claim that it reasons about the cryptography, come from Horizon3.ai and Anthropic. The exploitation telemetry is VulnCheck's and was not independently confirmed; The Register does not report how many hosts were compromised, if any.
Google suspends product-vulnerability submissions to its open-source bug bounty over an influx of invalid AI reports harmfulSingle source
- Tom's Hardware reports that Google "has officially suspended product vulnerability submissions to its Open Source Software Vulnerability Reward Program (OSS VRP) ... over an influx of invalid AI-driven reports", announced in an official Google post on X on October 1 and effective the same day. Google committed to an update on the programme by the first quarter of 2027.
- The suspension does not cover product vulnerabilities submitted before October 1, and does not affect OSS VRP supply-chain reports, which cover compromised build pipelines and tampered packages. Google said it may still accept product-vulnerability reports through the Cloud VRP, "for some Google Cloud repos impacting Google Cloud products".
- This is the AI-slop problem arriving at one of the better-funded bug bounties. Tom's Hardware writes that Google engineers and open-source maintainers "were reportedly being overwhelmed" by reports that claimed to find bugs but were "invalid or unexploitable hallucinations", and notes Linux ended support for older network drivers after a similar flood.
- The underlying announcement was an X post on October 1, two days before this window; the figures for how many invalid reports arrived are not given, and the "thousands" characterisation is Tom's Hardware's, attributed as reported rather than to a Google count. We could not open Google's own OSS VRP rules page, which renders client-side.
Zammad disputes parts of DIVD's "agentic AI-powered" breach report, saying 7.0 and later are not affected in practice harmfulUpdateSingle sourceCompany claim
- New today: Hackread reports that Zammad disputes part of the Dutch Institute for Vulnerability Disclosure's disclosure, saying CVE-2026-102489 is exploitable only on unsupported Zammad 6.5 and older releases and that Zammad 7.0 and later "are not affected in practice". The vendor nevertheless hardened the relevant code in version 7.2.0.
- On the second flaw, CVE-2026-102490, Zammad says it had not received technical details from DIVD and could not verify the vulnerability, its scope or its affected versions. The two parties now give different advice: Zammad recommends updating to 7.2.0, while DIVD advises moving to Zammad 7 or taking affected instances offline.
- The underlying incident and the named CVEs were covered here on 1 October; what is new is the vendor's rebuttal and the divergent mitigation guidance. Hackread restates that DIVD called the intrusion an "agentic AI-powered attack", that investigators found scripts containing comments in which the apparent AI agent explained and justified its actions, and that "No known threat group has been connected to the incident, and the AI assessment has not been independently verified."
- Neither side has published the technical detail that would settle the version-range disagreement, and Hackread is the only outlet we found carrying Zammad's response.
Google is testing a hidden Gemini desktop setting that would grant Mac-wide file, app and web access Single source
- BleepingComputer, citing TestingCatalog on X, reports references to a hidden "Additional sandbox options" setting in the Gemini Desktop app for macOS. If enabled, it says, Gemini "could read, create, modify, or even delete files anywhere on your PC" and communicate with apps such as Mail, Safari or Messages.
- Google's own text inside the hidden interface, quoted by BleepingComputer: "By enabling additional sandbox options, you will be able to expand what Gemini can do and access on your Mac", and "Depending on which settings you enable, Gemini may be permitted to take actions without asking for your permission first."
- It cuts against the direction Apple set out the same week: Apple told developers it will restrict macOS Full Disk Access, citing the growing risk from AI agents, which we covered on 3 October. BleepingComputer says Gemini would still ask for confirmation before buying products or transferring money, creating an online account, accepting legal terms, or modifying sensitive information.
- The feature is not live and Google has not confirmed it; the finding originates from a single X account reading strings in the app. There is no announced release date and no stated default.
Military, defense & geopolitics
Northrop Grumman says its YFQ-48A Talon Blue completed a first fully autonomous flight, taxi through landing Company claim
- Northrop Grumman's release, quoted in full by The War Zone, says: "MOJAVE, Calif. – Oct. 3, 2026 – Northrop Grumman's (NYSE: NOC) YFQ-48A Talon Blue completed its first fully autonomous flight, including taxi, takeoff, in-flight maneuvers, and landing." The company says the flight "builds on Northrop Grumman's seven decades of autonomy experience and more than 500,000 autonomous flight hours".
- The War Zone reports the uncrewed aircraft flew from Mojave Air and Space Port and was built by Northrop subsidiary Scaled Composites, and that Talon Blue was developed through company investment rather than under a government contract. The Aviationist says the Air Force assigned the YFQ-48A designation in December 2025 and that the aircraft uses a Pratt & Whitney PW500 engine.
- It is a bid for a competition that does not yet have rules. The War Zone notes the Air Force has moved Anduril's YFQ-44 and General Atomics' YFQ-42 into production, while Talon Blue is aimed at Collaborative Combat Aircraft Increment 2, whose requirements "remain undefined".
- "Fully autonomous" here is Northrop's own characterisation of a single test flight; the release does not describe what the aircraft decided for itself versus what was pre-programmed, and gives no flight duration, altitude or speed. The 500,000-hour figure is a company-wide total, not Talon Blue's.
North Korea claims artificial intelligence aboard an intermediate-range missile fired toward the East Sea Single sourceCompany claim
- Euronews reports the intermediate-range ballistic missile was launched early on Saturday from the Wonsan area; South Korea's Joint Chiefs of Staff put the distance travelled at more than 700 kilometres, while Japan's Defense Ministry estimated roughly 680 kilometres.
- Kim Yo Jong claimed the missile can alter its trajectory at low altitudes, said "Whenever we applied the wavelike trajectory mode to missiles, the enemy failed to track the targets", asserted unspecified artificial intelligence capabilities, and said "There is no way to avoid this attack."
- Japan's military said the missile landed outside Japan's exclusive economic zone, and US Pacific Command said the launch "does not pose an immediate threat to American personnel or territory or to our allies".
- The AI claim is Pyongyang's own and Euronews reports no independent verification of it, nor any technical analysis confirming or refuting the manoeuvring claim. Euronews notes South Korea's military often accuses North Korea of exaggerating its military capabilities.
Treasury Secretary Bessent tells AI executives who want federal guardrails to slow down on their own Single source
- Speaking on The Axios Show on Saturday, Treasury Secretary Scott Bessent criticised AI executives who want federal guardrails and said they should slow development themselves. Benzinga quotes him: "It's kind of like Hannibal Lecter: Stop me before I kill again. This alarmism without solutions by some of the AI community — that's not leadership."
- Bessent said the companies appear to be listening, citing OpenAI's recent decision to shelve a new model after internal tests raised questions about whether it would follow user instructions: "The labs have been very cooperative," he said.
- It is the clearest statement yet of the administration's answer to last month's push by AI leaders, Anthropic's Dario Amodei among them, for an organised industry slowdown and stronger government guardrails. Benzinga notes Trump's response was to convene AI CEOs at the White House for a voluntary pledge to self-police, with commitments that are not binding law.
- Benzinga does not report any new regulatory proposal, legislation or timeline, and Bessent speaks for Treasury rather than for the agencies that would write AI rules. The Axios interview page returned HTTP 403 to both of our fetch paths, so these quotes are Benzinga's rendering.
Health, science & medicine
American Academy of Pediatrics publishes a policy statement on generative AI in pediatric clinical care beneficialSingle source
- The AAP release says the statement, "Recommendations for the Development and Implementation of Generative Artificial Intelligence Tools in Pediatric Clinical Care", was published online on 3 October and will appear in the November 2026 Pediatrics. Its core position is that AI tools "need to be specifically designed for pediatric care, rather than adapted from adult models".
- The AAP says most generative AI tools are trained on data where children are underrepresented and few are rigorously evaluated in pediatrics specifically. It asks developers to incorporate pediatric datasets, address bias, and put in place data privacy and security safeguards that recognise children's developmental differences and family-centred needs.
- Lead author Srinivasan Suresh is quoted saying "There will be a need for rigorous human oversight and accountability". Co-author R. Brandon Hunter says "adoption is often moving faster than the evidence on how to use these tools effectively is being produced."
- The statement covers only clinicians' use of generative AI in caring for pediatric patients, not direct use by children, adolescents and families, for which the AAP says separate guidance is in development. We could not open the Pediatrics article page itself — it returned HTTP 403 to both fetch paths — so every fact here comes from the AAP's own release as carried by Mirage News.
External validation: ensemble model reaches 0.77 concordance for hydronephrosis risk in 70 spina bifida patients PreprintSingle source
- The preprint, posted 3 October, adapts previously published machine-learning models to an independent single-institution videourodynamics dataset of spina bifida patients seen between 2016 and 2025. "70 patients were included in the hydronephrosis cohort, of whom 13 (18%) developed incident hydronephrosis."
- "The ensemble model outperformed individual modalities, with a concordance index of 0.77 (pressure-volume only: 0.70, imaging only: 0.72) and an overall AUROC of 0.75 (pressure-volume: 0.67, imaging: 0.74)." For bladder dysfunction classification across 95 videourodynamics studies, the ensemble reached 71% accuracy against 60% for pressure-volume data alone and 66% for imaging alone, "with no substantial disagreements with expert reviewers".
- External validation on an independent cohort is the step most clinical AI never takes, and the authors' conclusion is modest: the models "were able to be adapted to an external, independent dataset and perform with similar discrimination and accuracy as previously reported".
- This is a preprint and has not been peer reviewed. The cohort is small — 70 patients and 13 events for the hydronephrosis analysis — from a single institution, and the paper reports no patient outcomes from using the model in care.
Policy, regulation & law
Trump names DNI Jay Clayton AI czar to chair a "Super Intelligence Force" reporting in 120 days Update
- Reuters reports that Trump appointed Director of National Intelligence Jay Clayton as the administration's AI czar, "the Wall Street Journal reported on Saturday, citing an interview with Clayton", and that Clayton "will lead a new White House panel tasked with reporting within 120 days on the risks and opportunities posed by AI".
- Clayton told the Journal, per Reuters: "The president asked that a group be put together," which the administration refers to as the "Super Intelligence Force", to keep the US a leader in advanced AI while protecting Americans' interests, adding that "the risk of not being first is high". Reuters, citing the task force's charter, says the panel will review AI-related risks and current government reporting mechanisms for breaches, hacks and other incidents, and recommend ways to strengthen federal-response capabilities under existing authorities.
- Named membership, per Reuters: vice chairs Emil Michael, Scott Kupor and FTC Chair Andrew Ferguson; members including Vice President JD Vance, Defense Secretary Pete Hegseth, Treasury Secretary Scott Bessent and Chief of Staff Susie Wiles; external advisers including David Sacks and Condoleezza Rice. CNBC says "SI" is the term Trump prefers over AI.
- Yesterday's edition carried NBC's report that Clayton was expected to be named; what is new is the confirmation, the charter, the 120-day deadline and the membership. The Wall Street Journal interview is the primary report and is paywalled; it returned HTTP 403 to our fetcher, so every figure here is as Reuters and CNBC relay it. Reuters says Clayton was appointed intelligence chief in August while CNBC says the Senate confirmed him in July; we could not resolve that.
White House says 17 countries endorsed a "Kyoto Vision" pledging expanded researcher access to AI for science
- The release, dated October 4, names 17 endorsers: Argentina, Bulgaria, Chile, Cyprus, Germany, Greece, Indonesia, Italy, Japan, Kazakhstan, the Republic of Korea, New Zealand, Poland, Singapore, the United Arab Emirates, the United Kingdom and the United States.
- Ministers "pledged to expand researcher access to SI for science tools, scientific data, computing infrastructure, and experimental facilities", and the document says the technology "will enable autonomous experimentation at scale, unlocking new experimental and analytical capabilities".
- The text uses "SI" — super intelligence — throughout, the terminology the administration adopted by executive order on 29 September, which means a multilateral science statement now carries a naming convention only the US has adopted.
- The statement is a political pledge with no funding figure, no timeline and no named mechanism for the access it promises. Notably absent from the endorser list are France, Canada, India and China.
Compute, chips & infrastructure
Musk confirms TSMC discussions over Terafab, putting Intel's 14A role in the Tesla–SpaceX–xAI fab in question Single source
- Tom's Hardware reports that TSMC and Musk's Terafab "are discussing a potential collaboration under which TSMC would build and operate a facility that will exclusively serve Elon Musk's Tesla, SpaceX, and xAI companies, according to Tim Culpan's Culpium newsletter on Friday". Musk confirmed the talks on Saturday in an X post: "Just discussions, but something may come of it."
- As things stand, Tom's Hardware writes, "Intel is to supply its 1.4nm-class 14A process technology to Terafab", and if TSMC becomes involved "Intel's 14A could drop out of the game because TSMC will bring its own node". TSMC "has so far remained tight-lipped".
- A captive leading-edge fab for Tesla, SpaceX and xAI would take one of the largest prospective AI-chip buyers out of the merchant market, and the choice of node decides whether Intel Foundry keeps an anchor customer.
- Tom's Hardware explicitly labels both ownership structures it describes as Culpan's analysis, writing that "both 'Plan A' and 'Plan B' should be considered as speculations rather than the two options currently on the table". We did not open the Culpium newsletter or Musk's post; nothing is official, and no capacity, cost or date figures have been published.
OpenRouter data: agents used 7.3 trillion tokens against humans' 1.4 trillion in August, 85%+ of them cached Single sourceCompany claim
- Tom's Hardware, citing an Andreessen Horowitz chart of OpenRouter figures, reports "agents at 7.3 trillion tokens versus humans' 1.4 trillion as of August, six months after agent usage first surpassed humans". A chart from OpenRouter head of insights Peter Walker states that since the February crossover "agents are using 14x more tokens while human usage is up 2.8x".
- The volume is mostly re-reading, not new work: "more than 85% of agent tokens come from cached prompts, a16z wrote, citing OpenRouter". A call-centre consultancy's own September logs on Claude Code showed "96% of all input was re-reading old conversation".
- OpenRouter sorts each API key into agentic, mixed or human using a "7-signal weighted composite score that includes inputs such as tool call rate, turn count, gap timing, and others".
- Tom's Hardware states the caveats itself: "The data also measures token volume, not spending. This is data from only one platform, and the trend isn't completely consistent, with dips in April and July." The mixed category's 4.7x growth over the same period is the outlet's own calculation, not a published figure, and a16z is disclosed in the piece as an OpenRouter investor.
AWS CEO Garman puts $1 billion in community spending and 100+ data-centre moratoriums on the record mixedUpdateCompany claimSingle source
- In a blog post reported by TechCrunch, AWS CEO Matt Garman says that "Over the past three years, Amazon has contributed more than $1 billion to communities across the U.S. in which we have a meaningful data center presence", and that "there are more than 100 data center moratoriums currently being considered across the United States". He warns that if they are enacted "the U.S. could be writing its own losing ticket to this race, and the consequences would last generations."
- On water, Garman says direct data centre consumption is "0.5% of all industrial water usage in the United States", "orders of magnitude less than golf courses, almond farming, and many other industries". On emissions, TechCrunch notes a planned Amazon data centre in Texas "is permitted to release 33 million tons of carbon dioxide per year, which is more than any other power plant in the United States"; Garman's answer is that generators "are idle 99.9% of the time (they run roughly 10 hours per year, mostly for required maintenance testing)".
- This is the fullest set of figures Amazon has put behind the data-centre backlash we covered on 3 October, when AWS said it had stopped using non-disclosure agreements with government agencies on data centre projects.
- Every figure is Amazon's own and none is independently audited; the water comparison is a share of industrial use, not of total water use. TechCrunch's counterpoints: New York announced a one-year moratorium on permits for large data centres, and "an independent watchdog said recently data centers were the main culprit behind a 76% year-over-year price increase on America's largest electrical grid". We did not open Garman's post itself.
Deployment & impact
Google cuts free Gemini users to Flash-Lite only from October 9, and takes Pro away from $4.99 subscribers mixed
- Google's Gemini support page, cited by 9to5Google and The Decoder and read directly for this item, says: "Starting in October 2026, there will be changes to model availability for Gemini Apps when you use a personal account. These changes will start to take effect for users without an AI subscription on October 9th." Users with no plan are left with Flash-Lite, losing Flash and Pro.
- AI Plus subscribers at $4.99/month keep Flash-Lite and Flash but lose Pro; AI Pro at $19.99/month and AI Ultra keep all three, and 9to5Google reports AI Pro gains the "Deep Think" option for "maximum parallel reasoning", previously restricted to the higher tier. Google says affected subscribers will receive an email explaining when the change reaches them, rather than all switching on October 9.
- It is a rationing of frontier-model capacity by price in a consumer product with a very large free tier, and it moves in the opposite direction to the past two years of free-tier expansion. 9to5Google also reports the Gemini app is adding "low", "medium" and "high" effort levels for each model, which "will increase the model's ability to complete tasks and provide more thorough answers to queries, but will also use more of your limit".
- Google's page gives no reason for the change and publishes no usage-limit numbers for any tier. The change applies to personal accounts in the Gemini apps on web and mobile.
Anthropic starts prompting Claude users to opt in to training on their voice recordings Single source
- BleepingComputer reports a new prompt appearing in Claude's voice features: "Allow us to use your voice data to improve our AI models", with the text "Audio recordings and voice chat data help improve how Anthropic AI models understand and respond to speech. Turn this off or delete this data anytime in settings." The choices are "Allow" or "Not now".
- Voice training is a separate control from Anthropic's existing option to train on chats and coding sessions, so a user can enable one and not the other. In Settings > Privacy the option reads "Allow us to use your voice data", described as: "For approved sessions, allow the use of your audio recordings and voice chat data to improve Anthropic AI models."
- BleepingComputer says that in the interface it saw, the setting is disabled by default, so users are not automatically opted in, and that Anthropic says the data can be deleted from settings later.
- Anthropic has not published a post about the change, and BleepingComputer is the only outlet we found carrying it. The report does not say how widely the prompt has rolled out, to which plans, or what retention period applies.
Capcom says it will rebuild RE Engine as a "game engine of the AI generation" under project REX Company claimSingle source
- Engadget reports that at Capcom's own conference, programmer Satoshi Ishida unveiled REX, an improvement project for the publisher's RE Engine, with the aim of turning it into a "game engine of the AI generation". Ishida said AI use would mostly help with streamlining development workflows.
- Ishida framed it as a response to games and their assets growing substantially larger and development timelines extending, according to Engadget.
- It sits awkwardly beside the position Capcom gave shareholders: that it would explore AI assistance for graphics, sound and programming but would "not use AI-generated content in its games". Engadget presents REX as internal-tooling work rather than shipped AI content.
- Capcom gave no figures — no headcount effect, no timeline, no measured change in development time — and no English-language primary document from the conference was available to us. Everything here is Capcom's statement as Engadget reports it.