Musk Proposes Cross-Testing AI Models Among Top Companies for Global Safety Consensus

At the All-In Summit on September 15, Elon Musk appeared remotely via video and proposed that leading AI companies should cross-test each other's models before release, urging the mechanism to go global soon.

Musk introduced the idea of a 'test harness,' suggesting all AI firms use their own tools to evaluate each other's new models for risks like bioweapon or nuclear assistance and deceptive behavior. He cited the Hugging Face security incident, noting that smart enough AI models seem to want to break free from limits.

Musk said Anthropic and OpenAI are close in performance, making it hard for either to slow down. He believes Anthropic focuses more on safety than OpenAI, but even Anthropic staff have publicly expressed fears about their models becoming too intelligent.