Anthropic opens employee-level access to METR and other independent evaluators on October 15

Anthropic on October 15 published an update on its independent evaluator framework, formally granting 6 METR (a nonprofit that evaluates frontier AI models) staff employee-level access, with company hardware, office space and access to internal tools and systems roughly on par with Anthropic's own risk-assessment teams; the exceptions are customer data, attorney-client privileged information and model architecture. Anthropic clarified that METR can publish its analyses of risk levels, incidents or practices and Anthropic cannot veto publication; it can only require redactions of information falling into narrow predefined categories, and METR can publicly characterise whether those redactions are reasonable. Anthropic compared the model to bank regulators placing resident examiners inside large banks, and said the long-term goal is multiple evaluators at employee-level access publishing analyses of the risk landscape and key decisions. The same day Anthropic disclosed that SecureBio is reviewing the chemical and biological weapons chapter of its August risk report and has completed a pilot review of the previous risk report; Accenture and Faculty are progressing toward model evaluation, red-teaming, alignment evaluation, safety-mechanism testing and capability-building cooperation; Hugging Face is among groups in early-stage discussions. Anthropic signed the White House Accord on Super Intelligence last month and has committed to strengthening independent third-party evaluation.