A Millennium Prize proof, and six Chinese labs named

OpenAI published a proof that smooth three-dimensional fluid motion can break down in finite time, resolving a Millennium Prize question open for roughly 90 years. It credits an unreleased internal model it calls significantly more capable than GPT-6 Astra, run as about 10,000 coordinating agents. NYU's Tristan Buckmaster, who published a related blowup result a day earlier with Anthropic's Levent Alpöge, says the pair drafted inside OpenAI's Codex and got no answer when he asked whether the model had trained on those sessions. OpenAI says no specific user data was accessed, while conceding it cannot rule out that de-identified usage data helped. Neither proof has been refereed.

The National Security Agency, the Federal Bureau of Investigation and the Cybersecurity and Infrastructure Security Agency issued a joint advisory saying DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.ai have extracted billions of tokens from Claude, GPT, Gemini and Grok since late 2024, likely with Chinese government awareness. That escalates a July accusation from the White House technology chief, which named Moonshot alone. Distillation, training a smaller model on a larger one's outputs, is a standard technique — but the agencies say it forms the core rather than a supplement of these labs' development. They also reject the number behind the 2025 market shock, DeepSeek's widely cited training cost of 5.6 million dollars. The advisory says it leaves out the cost of the data taken this way. China's foreign ministry calls the claims groundless.

Jacob Coxon, who worked at Anthropic and previously at OpenAI, resigned with a public warning that both labs are "gambling with our lives" and racing toward self-improving superintelligence. The unusual part is the response: Evan Hubinger, Anthropic's staff lead on alignment, agreed in public — saying the company does earnestly believe AI could kill all humans, putting the odds above ten percent within a decade, and adding that there is no plan yet for keeping a superintelligent system aligned. Senator Bernie Sanders has announced a bill that would bar firms from developing superintelligence.

Google DeepMind released AlphaGenome Atlas, running its AlphaGenome model across all nine billion possible single-letter changes in the human genome and publishing the predictions as a one-petabyte dataset, more than 30 times the size of the AlphaFold database. It is free for non-commercial use, with commercial access on Google Cloud to follow. Early collaborators, including the Broad Institute, Boston Children's Hospital and Memorial Sloan Kettering, report finding 22 percent more non-coding genetic associations and pinpointing a variant in the DNM1 gene tied to a severe childhood epilepsy.

Meta introduced Muse, a US-only personal agent that connects to a user's email, calendar, payments, health, smart home and shopping accounts and then acts on them, booking travel, filling forms and checking out through Link by Stripe. It runs on Muse Spark, Meta's frontier model — the agent and the model share the Muse name but are separate products. Muse is free with paid tiers at 20 and 100 dollars a month, and a payment card is required to start. It arrives less than two weeks after Meta settled multistate social media claims for 18 billion dollars, which is the trust question the launch has to answer.

Samsung and Taiwan Semiconductor Manufacturing said they will adopt ASML's high numerical aperture extreme ultraviolet lithography systems, which cost up to 400 million dollars each, for memory chips in 2028 and leading-edge logic in 2030. Intel already runs them in volume. More consequentially, all three have agreed with ASML to move the industry standard photomask from six inches to twelve inches, a change ASML's chief technology officer says could lift the machines' productivity by 40 percent. SK Hynix is weighing joining the consortium. ASML remains blocked from selling these systems to China.

Cognition, the company behind the Devin coding agent, raised two billion dollars, led by Andreessen Horowitz, at a 48 billion dollar valuation — nearly double the 26 billion it carried in May. Annualized revenue has gone from 492 million dollars in May to about 900 million now. TechCrunch's reading is that investors no longer treat AI coding as winner-take-all: the same firm backed Cursor before this, and Cognition now commands a richer revenue multiple than Cursor did at its own 50 billion dollar valuation in the spring.

An expanded class action says Anthropic deceptively marketed the usage multipliers on its Max plan, which runs 100 or 200 dollars a month above the 20 dollar Pro tier. The advertised five-times and twenty-times figures apply only within rolling five-hour windows that sit beneath a separate weekly cap, and reaching that detail takes two hyperlinks plus a trip to the Pro plan page. Anthropic argued in an earlier motion that this was the digital equivalent of flipping a product over to read the back label. The plaintiffs' attorneys are two former Federal Trade Commission lawyers.