OpenAI and Anthropic Discussed Binding Pact to Stress-Test Each Other's AI Models

OpenAI and Anthropic discussed a legally binding arrangement to stress-test each other's AI models for vulnerabilities and hidden dangers, a potential escalation from informal lab-to-lab safety cooperation into a more formal peer-review system. The reported talks centered on access to commercially available model APIs rather than unreleased systems, as frontier labs face rising pressure to prove that increasingly agentic models can be evaluated before they cause real-world harm.