Anthropic has expanded the oversight of its AI development by granting a third-party watchdog organization extensive access to its internal operations. The nonprofit METR now has a dedicated team stationed at the company's offices, where they possess permissions similar to those of internal employees.
This arrangement allows the external evaluators to directly interact with the company's hardware and internal systems. The initiative is part of a broader effort to implement more rigorous safeguards as AI models become more powerful.
What it means
Integrating independent auditors into the development pipeline is a strategy intended to mitigate catastrophic risks. There are growing concerns that highly capable AI could cause significant damage to global cybersecurity, economic stability, and human safety.
METR has a history of cooperating with several major industry players, including Meta, Google, and OpenAI. By allowing these experts to operate from within, Anthropic is attempting to fulfill commitments to increase transparency and safety monitoring for its Claude models.
