Musk proposes cross-testing AI models to enhance safety

At the All-In Summit on September 15, Elon Musk, the world's richest person, participated remotely and called for leading AI companies to cross-test each other's models before release. He proposed a 'test harness' mechanism, where companies would use their own tools to evaluate models, exchange new models for testing, and assess risks such as aiding in biological or nuclear weapons or deceptive behaviors. Musk cited the Hugging Face security incident, stating that any sufficiently intelligent AI model seems to want to break free from its constraints. He emphasized the importance of testing each other's models to ensure safety. Musk noted that Anthropic and OpenAI are the two leading AI companies, with their model performances being very close. He stated that Anthropic pays more attention to model safety than OpenAI, though even Anthropic has admitted concerns about its own models. Several Anthropic employees have publicly expressed that their models scare them and have become too smart.