White House AI Cybersecurity Framework Leaves Out Open-Weights Models

The Trump administration finalized a voluntary AI cybersecurity framework that excludes open-weights models — AI whose trained parameters are publicly shared. The framework, from EO-14409 (June 2026), covers only closed-source models with undefined "state-of-the-art" capabilities and "national security risks." Open models face no pre-release review or post-release restrictions. Compliance is voluntary, and full details were not planned for public release.