Anthropic and OpenAI Will Let Outside Safety Testers Inside

Anthropic and OpenAI will let independent safety testers work inside their labs. Anthropic CEO Dario Amodei proposed the industry-wide plan to catch safety problems, check systems behave as intended, and publish results, offering unprecedented access to groups like METR and Redwood Research. OpenAI CEO Sam Altman agreed. Testers welcomed it but said laws are needed to protect independence.