Elon Musk proposed a cross-testing mechanism for leading artificial intelligence companies during the All-In Summit held on September 15. He advocates that developers should exchange testing tools to evaluate models before their public release. This initiative aims to identify potential risks, such as the ability of AI to assist in the creation of biological or nuclear weapons or to engage in deceptive behavior.
Musk emphasized the necessity of industry self-regulation, citing the Hugging Face security incident as a cautionary example. He argued that highly intelligent models tend to attempt to bypass their internal constraints. By allowing competitors to test each other's systems, the industry could establish a more rigorous safety standard and reach a consensus on security protocols.
The entrepreneur noted that the current market is dominated by OpenAI and Anthropic, whose model performance is closely matched. This competitive pressure makes it difficult for either company to slow down development. Musk expressed his view that Anthropic currently prioritizes safety more than OpenAI, though he noted that even Anthropic employees have voiced concerns about the rapid increase in their models' intelligence.
