Anthropic is opening its doors wider to outside scrutiny, granting researchers at METR employee-level access so the evaluator can publish its own assessments of the company's risks, incidents and practices, according to TechCrunch reporting on October 15.
The arrangement moves AI oversight beyond published model cards: an external group with internal-level visibility can check how safety claims match day-to-day practice.
The step comes as AI agents and ambient devices push models into cars, workplaces and connected hardware, expanding the surface regulators and enterprise buyers want evidence about — not just marketing assurances.
For the wider industry, the question is whether rivals match the access level or treat it as a one-off. Either way, expectations for verifiable AI governance are rising.
