As cited
Copy frozen at (site build).
ai security
Anthropic Gives AI Safety Evaluators Insider Access
Anthropic granted the independent evaluator Metr employee-level access to assess its unreleased models before deployment. This access level aligns with Metr's recommended practices for evaluators to understand the behavior of highly capable artificial intelligence (AI) systems.
Why it matters: Security and AI risk practitioners need to know that Anthropic is enabling third-party evaluation of frontier models pre-release, which affects how AI safety risks are identified and disclosed before these systems reach users.
- First seen by Cybersecurity Tracker