Dario Amodei says independent AI safety reviewers should get "ongoing, employee-like access"

Dario Amodei of Anthropic has proposed that major AI companies allow independent safety reviewers to operate inside them on a continuing basis, with offices, credentials, and company computers so these teams can observe how safeguards are handled. The idea appears in his recent essay about curbing the field’s advance, amid a debate that followed an Anthropic researcher’s resignation and concerns about dangers to humanity. Amodei wrote that Anthropic is adopting the step on its own, and OpenAI’s Sam Altman praised it on social media and said his company would follow. The proposal is part of a broader industry discussion about whether AI development should be slowed, though experts question whether international coordination is realistic and critics worry outside evaluators may not be truly independent.