OpenAI, Anthropic, and Google Discuss AI Security

OpenAI, Anthropic, and Google DeepMind have been discussing AI system safety and the idea of a standardization body for several weeks. These discussions face antitrust concerns and divergent political positions in the United States, while OpenAI supports a provision of the FRONTIER Act favoring independent audits.

A Governance Framework Discussed Without Guaranteed Government Support

OpenAI, Anthropic, and Google are working together to create a standardization body for the AI industry. Sam Altman reportedly told his staff that this goal should be achieved without the support of the U.S. government. In July, Demis Hassabis called on the United States to establish a new body to act as a guardian, empowered to review the most advanced models and coordinate sector slowdowns in case of dangers. Some, including Sam Altman, have noted that inter-company discussions could expose participants to antitrust risks if the coordination is seen as suppressing competition. Politically, President Trump dismissed security concerns as a hoax and opposed regulatory tightening, arguing that a slowdown would benefit China. His chief AI advisor, David Sacks, stated that fears of existential risk were exaggerated.

Ongoing Technical Discussions on AI Safety

On Tuesday, Chris Lehane announced that the three companies had engaged in discussions for several weeks regarding AI safety. He also stated that OpenAI has been collaborating on this topic for several weeks with Anthropic and Google DeepMind. The previous week, Sam Altman hinted that he had private exchanges with other AI leaders. Additionally, Chris Lehane is reportedly in Washington to work with U.S. lawmakers on the catastrophic risks associated with these technologies.

Calls for Slowdown, Third-Party Assessments, and the FRONTIER Act

Dario Amodei called on the industry, in an essay published on Saturday, to collaborate to slow the pace of advancements in cutting-edge AI to avoid catastrophic risks. Sam Altman, Demis Hassabis, and Elon Musk supported this call. Sam Altman further stated that OpenAI would join Anthropic in integrating third-party assessors to monitor safety. Dario Amodei's essay proposed a narrow government exemption to allow for security-specific coordination, but Chris Lehane reportedly believed that companies did not need it. During the same meeting on Tuesday, Chris Lehane added that OpenAI supported a provision of the FRONTIER Act that would require leading labs to allow independent verification organizations to intervene to ensure that models are developed safely.