Joshua Achiam is calling for independent safety checks on AI developers. His October 3 response to David Robinson’s criticism also urges more reliable safety engineering and early, candid incident disclosure.
Achiam agrees that established safety practices from other industries have not taken enough root in AI. He also expresses greater confidence in his former OpenAI colleagues than Robinson does.
External verification should be free of conflicts of interest. Achiam argues that fixing problems as they arise is insufficient: OpenAI must earn trust that it can safely pursue more capable systems.
Miles Brundage also criticized deployment practices on October 3. He clarified that he still supports learning from experience; his objection concerns shipping already-known flaws.