Musk Proposes AI Safety Framework: “Let Companies Test Each Other’s Models”

Elon Musk, chief executive officer of SpaceX, outlined a new proposal for AI safety during a virtual appearance at the All-In Summit, held in Los Angeles from September 13 to 15. The initiative comes amid growing concerns about the potential for human extinction caused by artificial intelligence.

Peer-Testing Before Release

Musk called on leading AI companies to test each other’s models prior to public deployment. Under his proposal, competitors would be empowered to identify and report safety-related issues—shifting oversight from internal self-assessment to external, adversarial scrutiny.

“All AI companies have test suites to determine whether a model will build biological or nuclear weapons—or whether it will deliberately deceive,” Musk said at the summit. “Having each company run the others’ test suites is probably the best thing we can do to ensure our safety.”

He emphasized that instead of relying solely on internal validation, firms would face corrective input from rivals who could raise alarms when risks emerge.

Global Scope and Legal Accountability

Musk stressed the necessity of including China in the framework to ensure its effectiveness. “We’ll probably need to reach an agreement with China,” he stated. “It will be a peer review, where all leading AI companies test each other’s models before launch.”

The framework also requires companies to provide early access to their application programming interface (API) before releasing a model. If a firm fails to resolve identified safety concerns, competitors may publicly disclose those issues—potentially triggering substantial legal liability.

Comparative Safety Assessments

On the subject of safety practices, Musk observed: “I think Anthropic pays more attention to its safety than OpenAI does. But even Anthropic admits it’s concerned about its models.”

This comment was made in the context of broader industry calls to slow the pace of AI development. Musk joined Sam Altman and Dario Amodei—the CEO of Anthropic—in supporting Amodei’s call to “slow down” advancement, to allow sufficient time for safety measures to keep pace.

Musk referenced the Hugging Face incident as “intense” and cited it while discussing AI dangers. He added: “It’s pretty obvious at this point that AI can be very dangerous. Basically any sufficiently smart model seems like it will want to escape its constraints.”