The Trump administration plans early testing of new AI models to assess security risks and strengthen oversight before public release.