Anthropic gives outside AI watchdog “employee-like” access

Anthropic gave six METR evaluators employee-like access for independent AI safety review ahead of its IPO.

Anthropic has onboarded six employees of evaluator METR with company hardware, office space, and internal-tool access roughly equal to its own risk teams. Anthropic is not paying METR, which does not take AI-company funding and may publish findings without a company veto, though Anthropic can require redaction of customer data, privileged material, and model-architecture details. The arrangement comes as Anthropic prepares an IPO expected this year at a valuation above $2 trillion and briefs investors on safety. Anthropic continues working with Accenture on safety testing and with SecureBio on chemical and biological weapon risk.

• Six METR employees received hardware, offices, and internal access comparable to Anthropic risk teams.
• METR is unpaid, rejects AI-company funding, and can publish without an Anthropic veto.
• Anthropic may require redaction of customer data, privileged material, and model architectures.
• The move accompanies IPO preparations plus Accenture and SecureBio safety partnerships.