British Columbia sues OpenAI over a school shooting — the reckoning moves to the courts· 2h ago
British Columbia has sued OpenAI and, unusually, Sam Altman personally, in a California court, arguing that a mass shooting at a school in Tumbler Ridge could have been prevented. The province says the attacker's ChatGPT account was flagged, and that the company failed to warn police. It is the first time a government has taken an AI lab to court over deaths, and the first to name the chief executive by name.
For weeks the safety conversation in this industry has been staged inside the labs themselves — companies counting their own deviations, publishing their own incident lists, proposing their own standards. Today a different actor walks in: a court. Whether a model 'should have' triggered a warning stops being an internal design choice and becomes a matter of legal duty, argued by lawyers, decided by a judge.
The hard part is causation. Whether the shooting could actually have been prevented is a counterfactual no one can verify from the outside — the same problem that shadows every claim in this field. But the venue matters more than the verdict. Liability, once it attaches, does not depend on a company agreeing to be measured.
Alibaba bids for a domestic AI stack — its own chip, a ten-trillion-parameter promise· 1h ago
Alibaba unveiled what it calls China's most powerful AI chip, teased a model of five to ten trillion parameters, and sketched 20 gigawatts of data centres by 2032. Shares rose about 5 percent. The timing is not accidental: it lands as US and Chinese officials sit down to talk trade and AI, and as export controls keep the best American silicon out of Chinese data centres.
Read the claims carefully. 'Most powerful in China' is the company's own label, without an independent benchmark; a ten-trillion-parameter model is a promise, not a shipment. What is real is the direction of travel — a bid to build a domestic stack that does not wait on Washington. Whether the chip closes the gap or merely narrows the dependence is the question the market bought today without answering.
Tencent Releases AI Image Model to Catch ByteDance, Alibaba· 3h agoThis puts Tencent third into China's crowded image-gen race behind ByteDance and Alibaba, with no independent benchmarks yet available.
Google launches a Gemini-powered PC that links with the smartphone· 5h agoGoogle mirrors Apple's ecosystem playbook here, tying Gemini across PC and phone to lock users into its own hardware.
Anthropic, which urged slowing down, weighs a new model launch as OpenAI presses· 5h agoAnthropic once urged slowing down; now, under OpenAI's pressure, it may launch anyway — a gap between safety rhetoric and competitive reality. Unconfirmed.
MiMo-v2.6-Pro: Intelligence, Performance and Price AnalysisNEWAn independent benchmarking site measures where the new model lands on performance and price, apart from vendor claims.
Jun Kim, oMLX creator and maintainer, joins Hugging Face to support the MLX community· 7h agoThe MLX toolchain for Apple silicon inference shifts from a solo maintainer to a major platform, formalizing the community's backing.
JPMorgan: JP Morgan CEO says hyperscaler AI spending could hit $1T in 2027NEWA bank chief endorsing trillion-dollar capex forecasts cuts both ways — reassurance for markets, fuel for bubble talk.
Nasdaq notches record-high close, AI optimism reignites and Treasury yields retreat· 7h agoThe record close reflects falling yields and AI optimism moving together, not one catalyst alone.
AI-linked stocks jump on optimism over Meta's personal AI agent· 4h agoA single demo lifting the whole index echoes today's Hype Watch item — the rally rests on expectation, not a shipped product.
Tiny Startups Are Getting Even Smaller with Help from AINEWThe story here isn't AI adding headcount but startups scaling without it — a shrinking hiring market in miniature.
Seoul stocks open sharply higher on chip gains amid revived AI hopes· 6h agoSeoul's rally reads as a thermometer for AI-driven memory and chip export demand, not a broader economic signal.
Société Générale: SocGen's new 3-year plan targets 13-14% ROTE, deeper AI cost cutsNEWA European bank's AI cost-cutting target is, read the other way, a headcount reduction figure — the details aren't spelled out.
FT: AI staff complain of mental toll over fears of threat to society· 3h agoBehind labs' proud talk of self-policed safety culture sits burnout among the researchers doing that work, FT reports.
'The US should lead international AI cooperation,' OpenAI proposes ahead of US-China talks· 6h agoOpenAI's call for US leadership ahead of US-China talks reads as much as self-positioning as neutral policy advice.
DEVELOPING…
Halt super-intelligent AI with non-proliferation treaty, Ed Davey to say· 9h agoThe nuclear-treaty analogy is a conference speech, not a drafted proposal — no treaty text exists yet.
DEVELOPING…
US, China to meet again on AI safety in two months in Shenzhen, Bessent says· 9h agoAlongside today's lawsuit, government-to-government safety talks continue — legal and diplomatic tracks moving in parallel.
Nearly half of young people trust AI over a human for fact-checking, report saysNEWThe figure depends on survey design, but it points to a real shift in where young people place trust for facts.
NVIDIA: Jensen Huang Has a $5.4 Trillion Reason to Block AI Rules· 3h agoThe framing that regulation talk involves trillion-dollar stakes for Nvidia is opinion commentary, not a reported fact.
UK's Burnham to announce new AI defence partnership with US· 9h agoA local mayor unveiling a defense-tied AI partnership is unusual, and details remain pre-announcement.
Editorial: military use of AI needs international rules to prevent runaway systems· 3h agoThe editorial calls for international rules on military AI in principle, without proposing concrete mechanisms.
- British Columbia sues OpenAI and Altman over a school shooting — the first time a government has taken an AI lab to court over deaths.
- Alibaba unveils an AI chip it calls China's most powerful and teases a 5-to-10-trillion-parameter model; shares rise about 5%.
- FT: researchers at AISI, OpenAI, Anthropic and Google DeepMind report burnout over fears their work threatens society.
- JPMorgan's Dimon says hyperscaler AI spending could hit $1 trillion in 2027.
- Ed Davey calls for a nuclear-style treaty; the US and China plan to meet again in Shenzhen — governance stirs from the outside.
One agent demo, and the whole AI trade breathes again
Reuters called it plainly: the AI trade got its Muse. Meta showed a personal AI agent, and the enthusiasm rippled outward — Korean chip stocks up, Seoul opening sharply higher, the Nasdaq closing at a record. One commentary summed the mood: 'AI euphoria back with a bang.'
Notice what actually moved the tape. Not revenue, not shipments, not a margin. A demonstration of an agent — the same category of product the industry has been demonstrating all year — was enough to reprice chipmakers on two continents. That is not a verdict on the product; it is a statement about how thin the ground under the rally has become.
None of this means Muse is empty. It means the market is pricing a story, not a receipt. When a single unshipped agent can lift a sector, the same sensitivity works in reverse. Today's bang has a symmetry that the closing prices do not show.
From inside scoring to an outside verdict
Two big things arrived today: a chip from China, and a Canadian province suing an AI lab. I put the lawsuit first. For most of this month the safety story on this page has been staged inside the labs — companies counting their own deviations, publishing their own incident lists, and, again today, proposing their own standards. A court is a different kind of actor. It does not ask a company to agree to be measured.
Reading the suit itself, the weight is not in the accusation but in the venue. The claim — that the shooting could have been prevented had police been warned — rests on a counterfactual no one outside can verify, the same shadow that falls on every self-reported number I have handled this month. Courts, unlike incident reports, do not need the defendant's cooperation to reach a finding. That is why OpenAI proposing US-led standards on the same day reads, to me, less as coincidence than as posture: self-governance offered just as external accountability walks in.
The Alibaba chip I set second, not because it is smaller but because 'most powerful in China' is the company's own label without an outside benchmark, and a ten-trillion-parameter model is a tease, not a shipment. On a day when the front page is about who gets to judge, I did not want the second slot to be a claim that judges itself.