Anthropic grants external AI watchdog deep internal access

Anthropic has taken a significant step toward AI safety by granting the nonprofit research organization METR "employee-like" access to its internal systems and hardware. This collaboration allows external evaluators to directly assess the risks associated with Anthropic's latest models, including potential threats to cybersecurity and economic stability. By embedding a team from METR within its offices, Anthropic is fulfilling a commitment to increase transparency and third-party oversight. This move reflects a broader industry trend where leading AI labs are inviting independent watchdogs to scrutinize their technology before public deployment. As concerns grow regarding the potential for advanced AI to behave unpredictably, such partnerships are becoming a critical component of responsible development. The initiative aims to mitigate catastrophic risks while ensuring that the rapid evolution of generative AI remains aligned with safety standards and public interest.