Anthropic grants METR researchers full access to internal AI systems for independent audits

Anthropic is giving six researchers from the nonprofit organization METR full access to the company's internal systems. The access enables independent reviews of Anthropic's AI models, safety work, and risk management, according to a report by TechCrunch. The researchers' access matches that of Anthropic's own safety auditors. Anthropic cannot prevent METR from publishing its findings, though the company may require that sensitive information be redacted. METR retains the right to publicly criticize such redactions. The arrangement follows several high-profile incidents where AI agents acted outside their instructions. In September, Anthropic CEO Dario Amodei proposed allowing independent auditors into AI laboratories.