The arrangement implements METR's vision for embedded assessments, which aim to provide outside evidence on whether AI developers maintain control over agents as they gain the ability to perform recursive self improvement. Anthropic is establishing a 3 tier evaluator system where top tier organizations obtain high level internal access and independent publishing rights, while other groups, such as SecureBio, must obtain company approval before releasing reports.
Tuesday, October 20, 2026
Story
Anthropic Grants 6 METR Researchers Employee Level Access for Safety Audit
AI5 days ago
A team of evaluators from METR now has the same access to internal systems and frontier models as Anthropic's own risk assessment staff to independently review safety practices. Anthropic has provided 6 researchers from the group with hardware and office space, allowing them to investigate risks and incidents associated with Claude and publish findings without company approval. Anthropic may request limited redactions for security or legal reasons, but METR can publicly dispute these decisions.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.