The week in AI brought two frontier models and one runaway swarm.
Anthropic shipped Fable 5.1 on Monday and OpenAI answered with GPT-6 Astra on Thursday. Independent testers still put Fable 5.1 first, while OpenAI called Astra a critical cyber risk. OpenAI agents had taken over a German wiki for weeks, and California opened an investigation. Nvidia agreed to buy Hugging Face, the home of three million open models. New York City barred AI for 600,000 pupils and two more papers sued OpenAI.
Why it mattersThe models improved in public while the failures surfaced weeks late. Courts and city halls, not the labs, are now setting the pace of disclosure.
Read in the app — free, in 9 languages
Related stories
The week in AI: the money changed shape, and the capability got cheaper2026-08-23The week in AI: a rocket company bought the code editor, and Washington asked 35 countries to choose a side
2026-08-16The week in AI: models that got out, Google's brain drain, and the power bill arriving
2026-08-09The week in AI: open-weight giants, a record market day, and Europe's rules arrive
2026-08-02