The People Who Make AI Will Let Outside Inspectors Check Their Work

Anthropic and OpenAI will let outside safety experts work inside their companies. Anthropic chief Dario Amodei suggested the idea so inspectors, like car testers, can spot dangers, check the AI behaves, and share findings. He offered unprecedented access to groups METR and Redwood Research. OpenAI chief Sam Altman agreed. Experts welcomed the plan but said laws should keep inspectors independent.