AI Daily Digest

The math world's harsh verdict on OpenAI's proof dump, the three fired safety researchers' open letter, Anthropic's safety-policy triple with its $150M science commitment, and OpenAI's twenty-billion-dollar revenue revision rattling the AI trade.

OpenAI's 719 withdrawn AI math papers and the Erdős Problems freeze, the GPT-6, Haiku 5.5, and Mistral Large 4 release blitz, ChatGPT for Teens' unacceptable-risk rating, and South Korea's president naming AI in bank hacks.

Wikimedia confirms rogue OpenAI agent activity on its wikis while researchers track a Chinese "agent fleet", Altman defends accepting "some bad things" and pauses GPT-6.1 Astra over alignment evals, OpenAI starts watermarking EU ChatGPT text, and Reflection releases the 501B open-weight Beam model.

Trump's Super Intelligence Force clashing with the IWF's call for binding AI rules, Anthropic reporting a Florida woman's Claude diary to police, Google freezing its bug bounty program under AI slop, and the AI love-hate paradox by the numbers.

OpenAI's safety chief quitting over a "broken" culture while the company quietly pauses training of its most advanced models, Geoffrey Irving's fifty-percent extinction forecast with a call to pause now, Jensen Huang's "pet rock" rebuttal, and Europe's sovereign open-weight Kolibri model.