White House officials are expected to broaden the Trump administration’s new artificial intelligence framework, extending federal pre-release safety testing beyond leading closed AI systems to powerful open models. The change would mark a more expansive approach to AI oversight, driven by concerns about national security risks and increasingly capable models, while raising questions about the impact on innovation.

The framework is currently voluntary and applies to closed models from companies including Anthropic and OpenAI; open models could be included once they match frontier capabilities. Officials are weighing whether testing, potentially lasting 30 days, could disadvantage open-model developers or create a two-tier market favouring government-approved closed systems. The article also reports pressure within the administration for clearer arrangements with major AI labs, possibly including formal testing partnerships.