Elon Musk called on xAI, OpenAI, Anthropic, Google, Meta and three or four leading Chinese AI companies to let competitors run safety tests on their models before public release. Speaking at the All-In Summit in Los Angeles on Monday, he proposed giving rivals access to run a “test harness” to evaluate safety.

Musk proposed one or two weeks of early access for competitors and informal calls weekly or every two weeks. He argued that rivals would recognize security risks and have an incentive to flag models whose release should be delayed. “Instead of grading your own homework you would at least have competitors grading your homework,” he said. He advocated government intervention if a company refused to reduce a serious danger.

OpenAI, Anthropic and Google have already been coordinating on AI safety for several weeks and have discussed creating a standards body focused on testing and auditing models. OpenAI says it does not need an antitrust waiver for safety coordination, while Anthropic CEO Dario Amodei has requested a narrow waiver to enable safety talks among rivals as part of his call to slow advanced AI development.