Demis Hassabis, co-founder and chief executive of Google DeepMind, proposed creating a US-led Frontier AI Standards Body to assess the most capable AI models before release. In a personal essay published on 14 July, Hassabis said the body could operate as a federally supervised public-private partnership or as a self-regulatory organisation, similar to the Financial Industry Regulatory Authority. It would be funded primarily by the AI industry but governed by a board including independent technical experts and representatives of the open-source community. The organisation would develop regularly updated benchmarks to determine whether a model qualifies as ‘Frontier-class’. Developers whose systems cross those capability thresholds would be designated ‘Frontier Labs’, regardless of where they are based or whether their models are open or closed. Smaller developers, start-ups and academic projects below the threshold would be exempt. During an initial voluntary phase, Frontier Labs would submit models for assessment up to 30 days before release. Once effective, passing the assessment could become a requirement for deploying frontier models in the US market. Evaluations would examine capabilities related to cybersecurity, biological threats and other high-risk areas, alongside signs of deception or attempts to bypass safety controls. Standards would be updated every quarter, and the body could work with federal agencies, US National Laboratories and third-party auditors. Hassabis presents the proposal as a response to intensifying commercial and geopolitical competition and the possibility of increasingly autonomous AI systems.