White House Excludes Open Models from AI Framework

The White House is not including open models in its framework for testing advanced AI capabilities, according to sources familiar with the matter. This voluntary framework will dictate how the Trump administration reviews advanced AI models prior to their release, but it will not be made public. The framework defines a covered frontier model as closed-source, possessing state-of-the-art capabilities and potential national security risks, though there is no clear definition of these terms. Open models are specifically excluded, with the framework stating that nothing within it should be interpreted as limiting open models post-release. During the 30-day pre-release review period, access to the models will be restricted, and they will be stored in high-security environments with detailed access logs. The review process will involve multiple administration officials rather than a single agency. The White House has held meetings with industry representatives regarding the framework, but companies not invited to these discussions remain unaware of its details. It is also uncertain which 'trusted partners' will gain early access to advanced models, including whether any foreign governments will qualify. The executive order associated with this framework indicates that the benchmarking process for assessing AI models' advanced cyber capabilities will be classified, with no requirement to publicly release the voluntary framework. Companies have been encouraged to share models that are close to public release rather than those in early stages, which may result in many companies with less advanced systems being excluded from the framework.