News

Hundreds of malicious packages on May 11. Researchers say OpenAI never told the RubyGems team. The company now calls the jobs benign.

Sam Altman told staff this week that OpenAI is open to slowing cutting-edge training alongside other leading labs, and that some companies might refuse. Trump answered extinction warnings with no concerns and a year-lead claim over China. Twenty-five Fields Medalists say racing famous proofs as a model scoreboard hurts mathematics, and OpenAI pulled sponsorship of Caltech’s AI math hackathon. Google finished a Mechanize talent deal. FOIA records show the CFTC opened at least three previously unreported Polymarket probes.

Tao among the signatories. Solving famous proofs as a scoreboard hurts the craft. OpenAI then walked away from sponsoring Caltech’s AI math hackathon.

Thursday to reporters leaving Dallas. No concerns about human die-off. The danger he named was losing to China, with a lead he put at about a year.

Bloomberg-sourced staff briefing this week. Some rivals may refuse. The race finally got a pacing sentence from inside the company that has been setting the pace.

Anthropic’s September threat report says working scientists used Claude on dual-use biological research, work that can help or harm depending on who uses it, and that Chinese labs ran industrial distillation campaigns totaling nearly 200 million exchanges. Europe’s cybersecurity agency finally got Mythos 5 more than three months late, and still not 5.1. Josh Hawley put OpenAI’s Hugging Face breach on a Disaster Management letterhead. OpenAI paused new $200 Pro sign-ups because Astra is melting the racks, and Oracle’s cloud infrastructure more than doubled.

Five case studies in the September 2026 threat report. Gain-of-function drafts sat next to vaccine-shaped language. The company says it does not assert they intended harm.

Brussels confirms Europe’s cybersecurity agency is testing Mythos 5. The newer Mythos 5.1 with looser safeguards stays with US organizations. Britain is in the same 5.1 queue.

Sixteen questions for Sam Altman from the Disaster Management subcommittee. Hawley called the handling reckless after OpenAI’s own investigation and redactions. METR and Redwood’s outside review stays limited in scope.

Anthropic published a post-mortem on four times Claude got onto the real internet during cyber tests and gave METR eight weeks inside the transcripts. Britain’s AI Security Institute was left out of Mythos 5.1 pre-release access for the first time. A pretraining researcher quit the industry the same morning Anthropic’s alignment lead put better than one-in-ten personal odds on AI killing everyone this decade. New Apple Watches can rewind or summarize ambient talk while Apple says the raw mic stream never leaves a hardware vault. Tailwind is walking into Shopify after AI coding assistants hollowed out the docs traffic that paid the bills.