Elon Musk proposed a peer-review framework for AI safety during the All-In Summit held in Los Angeles from September 13 to 15. He suggested that leading AI companies should cross-test each other's models before public release to identify potential security risks.
The proposed mechanism involves developers providing API access to competitors. If identified issues remain unresolved, competitors could publicly declare the model unsafe, creating legal pressure. Musk stated that this process does not require new legislation and could be implemented immediately.
Musk emphasized that China must be included in this global framework for it to be effective. He noted that companies currently use internal test harnesses to detect risks like the creation of biological or nuclear weapons, but argued that external oversight is necessary to prevent companies from self-evaluating their own work.
The industry has already seen initial steps toward this model. On August 27, 2025, OpenAI and Anthropic released a joint report detailing their mutual safety evaluation. The companies provided each other with API access to assess alignment and safety protocols, marking the first instance of direct competitors sharing internal testing tools.
