DeepMind Chief: US-Led Body Needed to Test Frontier AI Models

Google DeepMind's chief is calling for a U.S.-led body to test advanced AI models. This proposed organization would cover frontier models developed both in the U.S. and internationally, including both open and closed systems. Its focus would be on testing dangerous capabilities, cybersecurity risks, and potential bypasses of model safeguards. What's interesting is the idea of modeling this body after the Financial Industry Regulatory Authority, or FINRA, which oversees U.S. securities firms. The DeepMind chief believes a U.S.-initiated system could set international standards as AI systems become more capable. This framework aims to provide a common testing process for AI companies, avoiding separate licensing for every model. Independent testers could then compare models against shared thresholds and update tests as new capabilities emerge. The White House has already begun pre-release oversight. An executive order in June sought voluntary access to frontier models for up to 30 days before release to test advanced cyber capabilities. Several major tech companies, including Google DeepMind, Microsoft, and xAI, have agreed to provide models for federal national security testing. This proposal highlights a shift in the AI debate, moving from whether oversight is needed to who conducts the tests and how models are regulated.