Safety & policy

Anthropic lets safety watchdog METR inside its office

Anthropic has given METR, a nonprofit that has worked with OpenAI, Google and Meta, employee-like access to its hardware and internal systems. Six METR staff now work in the company's office with tools roughly on par with its own risk teams, Anthropic says. METR can publish its findings and Anthropic can't veto that, though it can require redaction of categories such as customer data. Critics warn of conflicts of interest and want tougher government regulation, while Anthropic compares the set-up to bank supervision.

Why it matters: If your business relies on Claude or other AI tools, ask vendors how independent evaluators check their safety claims.

Source: Ars Technica