Anthropic Grants Third‑Party Evaluators Employee‑Like Access to Its Systems

LATEST MATERIAL CHANGE

What changed?

Anthropic has taken a concrete step to strengthen oversight of its AI models by allowing the nonprofit METR to embed a team in its office with “employee‑like” access to hardware and internal systems. METR, which has worked with major labs such as OpenAI, Google, and Meta, will assess the risks of Claude’s capabilities and help identify safeguards against potential catastrophic outcomes. This move follows growing calls for greater transparency and external review as AI systems become more powerful and could impact the economy, cyber security, and broader society. By opening its internal environment to independent scrutiny, Anthropic aims to demonstrate its commitment to responsible AI development and mitigate the dangers of uncontrolled model behavior.

Why it matters

Providing external evaluators with deep access signals a shift toward greater accountability in AI research, helping to prevent risks that could threaten economic stability, cyber security, and human safety.