AI news, 9 September: OpenAI claims AI cracked a Millennium Prize math problem

OpenAI says AI solved the Navier-Stokes problem. On 8 September OpenAI announced that an unreleased internal model, working with roughly 10,000 AI agents, produced a solution to the Navier–Stokes existence and smoothness problem — one of the seven Clay Millennium Prize Problems — in about 88 hours. According to OpenAI and reports from the BBC, The Guardian and The New York Times, the proof shows that initially smooth fluid equations can develop a singularity (speeds becoming infinite) in finite time while keeping energy finite. The run involved millions of agent messages and huge compute costs estimated in the millions of dollars; the result was then formalized in the Lean proof assistant. If independently verified, it would mark a landmark for AI in pure mathematics. The claim has sparked controversy, however: mathematicians including Tristan Buckmaster say OpenAI accelerated after hearing of related work by Buckmaster and an Anthropic researcher, though OpenAI denies accessing unpublished material improperly. Verification by the wider math community remains essential.

Mistral raises €3 billion in Europe’s biggest tech round. French AI company Mistral AI closed a €3 billion ($3.5 billion) Series D led by Samsung Electronics, lifting its valuation above €21 billion — described by the firm and outlets including TechCrunch and CNBC as the largest equity fundraising ever completed by a privately held European technology company. Co-leads included the EU-backed Scaleup Europe Fund managed by EQT. The capital will expand research, compute capacity, infrastructure and international reach as Mistral positions itself as a sovereign, open-weight alternative to leading US labs.

Meta launches Muse personal AI agent. Meta released Muse in the United States, a personal AI agent that connects to users’ email, calendars, payments and other apps to handle real-world tasks such as booking travel, shopping, filling forms and sending messages. It runs inside a dedicated secure virtual machine and stresses user controls and privacy. A free tier is available alongside $20 and $100 monthly plans, according to TechCrunch, Bloomberg and Meta’s own announcement. The debut comes shortly after Meta’s large multi-state child-safety settlement.

US agencies warn of industrial-scale Chinese model distillation. The NSA, CISA and FBI issued a joint cybersecurity advisory accusing Chinese AI firms including DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI of “aggressive, malicious” knowledge-distillation campaigns against US frontier models (variants of Claude, GPT, Gemini and Grok) since late 2024. The agencies say the firms extracted billions of tokens across millions of exchanges, likely with government awareness, to shortcut their own development, according to the advisory and reporting by Reuters and Bloomberg. They urge US providers to improve detection of anomalous usage.

DeepMind maps effects of 9 billion DNA variants. Google DeepMind released AlphaGenome Atlas, a freely available (non-commercial) database that predicts the molecular impact of every possible single-letter change — about 9 billion variants — across the human genome. The roughly 1-petabyte resource, built with the AlphaGenome model, is intended to help researchers more quickly identify variants linked to disease and traits, according to Nature and DeepMind’s announcement.