White House AI Plan Spares Open Models, Raising Safety and Competition Questions

The White House is finalizing a new framework for vetting advanced AI models before release, but plans to largely exclude “open-source” or “open‑weight” systems from that process, according to multiple reports. The framework is designed for so‑called frontier models—high‑end, closed systems like those from OpenAI and Anthropic—whose capabilities and national‑security implications are deemed significant enough to warrant pre‑deployment cybersecurity and safety testing by the U.S. government.

Axios and other outlets report that the draft framework defines its scope narrowly around these proprietary models, while open models would generally fall outside the vetting regime. That carve‑out comes as open‑weight systems gain momentum in both U.S. and Chinese labs, prompting debate over whether they should really be treated as lower risk. Another point of controversy is transparency: the administration has briefed a small group of major tech firms on the rules but has not released the document publicly. Supporters frame the move as a way to manage extreme risks without stifling open innovation, while critics warn it could leave a growing class of powerful, widely distributed models in a regulatory blind spot.