ANTHROPIC OPENS DOORS: Outside Watchdog Gets 'Employee-Like' Access...

Anthropic embedded a team of six external evaluators from METR, a nonprofit AI safety organization, directly into its offices with access to internal systems and hardware—a move the company frames as parallel to bank regulators stationed inside financial institutions. The arrangement lets METR publish findings without Anthropic's approval, though the company can request redactions for sensitive material like customer data or model architecture details. The step responds to mounting pressure on AI labs to demonstrate stronger safety oversight as their systems grow more powerful, though critics worry third-party evaluators embedded within companies could face conflicts of interest similar to those that plagued financial regulation.