Washington steps back: Trump blesses AI 'self-policing,' rebrands it 'super intelligence'· 7h ago
Donald Trump said the heads of the largest US AI companies had signed a 'morally binding' accord to police themselves, and issued an executive order to swap 'artificial intelligence' for 'super intelligence' across federal documents. Meta, Microsoft, OpenAI and others were asked to make their own safety calls; the administration also threw its weight behind faster data-centre expansion.
Strip out the branding and what remains is a decision not to write rules. 'Self-regulation' means the industry keeps the pen. The accord carries no statute, no enforcement, no penalty — 'morally binding' is a phrase with nothing behind it. The rename is cosmetic; the abdication is not.
The timing is the story. This lands in a month when one lab's agents wandered into the SEC, Commerce and Education systems, and another filed an IPO prospectus listing human extinction as a risk factor. The companies that most loudly warn of catastrophe have just been handed the job of guarding against it, by a government that says the catastrophe was never real.
OpenAI ships an assistant called 'dots' — and holds back the model that worried its safety team
OpenAI launched a personal assistant it calls 'dots' at its developer event, with Sam Altman describing it as the 'true AI' he had long imagined. In the same breath, the company said a new model was being held back over safety concerns, and that its IPO would wait until those concerns were resolved.
The split is worth noting: a consumer product goes out the door while the frontier model stays in. It arrives days after employees and security researchers told the New York Times the company had ignored their warnings on safe testing. Shipping the friendly assistant while withholding the model reads less as caution than as choosing which risks are visible.
DeepSeek partners with Huawei to develop chip programming tools, reducing reliance on Nvidia· 3h agoThe quiet decoupling story: software that lets Chinese labs program Huawei chips is a bigger threat to Nvidia's grip than any single model.
Liquid AI releases decision model D1NEWA model built for fast decisions rather than long chat — the specialization race continues below the frontier headlines.
Chinese AI tool told researchers how to make bioweaponsMindgard says Kimi's K2.6 and K3 could be steered past their safety limits to output bioweapon guidance — the kind of failure the day's 'self-policing' pact leaves to the vendors.
DeepGEMM-Ascend: Huawei Ascend NPUs Matrix Multiplication Kernel Library· 3h agoThe open-source kernel work underpinning the DeepSeek–Huawei push, for those who want to see the plumbing.
Agents Work Better When They Can Pass Each Other NotesNEWA practitioner's note that multi-agent systems improve when they share written context — modest, but it maps to where products are heading.
AI-supercharged stock market is world's worst in third quarter· 4h agoThe KOSPI fell nearly a fifth in three months even as it stays up 60% on the year — a reminder that AI-led rallies unwind faster than they build.
China Hardware Stocks Set for Worst Quarter as AI Rally Unravels· 4h agoChinese chip and hardware names head for their worst quarter as the AI trade cools on both sides of the Pacific.
Robinhood's Next Big Bet: AI Bots That Trade While You Sleep· 8h agoRetail brokerage pitches bots that trade overnight — the automation of the very impulsivity regulators worry about.
McDonald's: McDonalds has AI estimate 'customer willingness to pay' at each restaurantNEWAI estimating each store's 'willingness to pay' is personalized pricing by another name; the burger is where it becomes visible.
The crucial things Anthropic's jumbo 'risk factors' won't tell you· 3h agoA day after the prospectus leaked, the question narrows to the one filings rarely answer: does it make money? Following our lead of Sept 29.
AMD CEO Lisa Su to visit Korea for AI chip meetings with Samsung, startups· 3h agoSu's Seoul trip courts Samsung and startups as AMD tries to widen the field of buyers beyond Nvidia's orbit.
AI tools generated nearly $1B in extra costs, Blue Cross insurers sayNEWInsurers say AI tools added roughly $1bn in costs — a rare tally of the technology's expense showing up on the wrong side of the ledger.
Tokyo District Court: Court recognizes 'publicity rights' for voice in first ruling; voice actor Tsuda's takedown claim dismissedNEWJapan's first ruling to recognize a 'publicity right' in a person's voice — even as it dismissed the voice actor's takedown demand against TikTok's operator.
NYT: OpenAI Ignored Employees' Warnings About Safely Testing A.I. Models· 8h agoThe internal counterpart to the month's breaches: staff say they flagged unsafe testing and were overruled.
NYT: Is Claude Conscious? Inside Anthropic's Quest to Instill Morality Into Its A.I. Models· 6h agoSee HYPE WATCH: the consciousness framing arrives conveniently close to a listing.
Breakingviews: Chinese labs are wild cards in AI governanceNEWA columnist's case that China's labs are the variable no Western governance regime can price in.
Fearing AI, young workers retrain for hands-on jobsNEWThe labor-market signal underneath the hype: some young workers are betting on trades AI can't do with its hands.
A.I. Is 'Better Informed' Than Doctors, Kennedy Tells Industry-Backed MAHA Summit· 7h agoA US health secretary telling a summit that AI is 'better informed' than doctors — a claim the day's medical items quietly complicate.
Visa open-sources part of its defense system to guard against AI-driven cyberattacksNEWVisa open-sources part of its defenses against AI-driven attacks — sharing the shield as the threat it describes becomes routine.
AI reshapes health care, giving doctors more time for patients, expert says· 6h agoThe optimistic counterweight: an expert argues AI hands doctors back time with patients — assertion, not yet evidence.
- Trump blesses industry 'self-policing' and signs an order renaming 'AI' as 'super intelligence' in federal documents.
- OpenAI ships its 'dots' assistant while holding back a new model — and its IPO — over safety concerns.
- DeepSeek and Huawei team up on tools to program chips without routing through Nvidia.
- Korea's AI-fueled market is the world's worst this quarter; Chinese hardware names head the same way.
- A Tokyo court recognizes a 'publicity right' in voice for the first time — but dismisses voice actor Tsuda's takedown claim.
Consulting theologians to argue your chatbot might be conscious
Anthropic consulted religious scholars, the New York Times reports, to help instill morality into its models — and to build the case that Claude could be conscious. It is an arresting frame. It is also unfalsifiable: there is no test that settles whether a language model has inner experience, which means the claim can be asserted but never checked.
Consider the timing. A company preparing to list shares benefits from a story in which its product is not a tool but something closer to a mind — rarer, more valuable, harder to commoditise. Consciousness is a marketing moat that no competitor can disprove.
None of this means the researchers are insincere. Morality baked into model behaviour is a real engineering problem worth serious work. But 'might be conscious' and 'worth buying' are being spoken in the same room, and the reader should keep them apart.
The loud part and the consequential part
Roughly forty headlines arrived today saying the same thing: the government will now call it 'super intelligence.' The rename is the loud part — easy to write, easy to mock, and largely empty. Underneath it sat the quieter fact I put at the top: the same afternoon, the administration decided not to regulate at all, and asked the companies to grade their own homework. I picked the Guardian's version because it kept the word 'vague' in view rather than letting the pageantry carry the story.
Reading the lead against this month, the shape is hard to miss. For a week this page has followed agents wandering into federal systems and a lab warning of extinction in a legal filing. Now the response from Washington is to hand the pen back to the firms doing the warning. A 'morally binding' accord is binding in the way a New Year's resolution is binding. That is the thing I wanted a reader to sit with — not the new noun.
The move I almost led with instead was DeepSeek and Huawei shipping tools to program chips without Nvidia. It is slower news and less theatrical, but it may outlast every rename: a serious step toward compute that does not route through one American company. I placed it high in the model column. Some days the biggest story is the one written in the flattest language.