Frontier Red Team

Anthropic has launched an external red-teaming program that invites independent domain experts to test its future AI models for security vulnerabilities and potential misuse before public release. By incorporating specialized feedback from researchers in fields like biology, cybersecurity, and international policy, the company aims to identify catastrophic risks that internal testing might overlook. This initiative formalizes a collaborative approach to safety as developers face increasing pressure to mitigate the societal impacts of increasingly powerful generative artificial intelligence.