Trump AI Framework Excludes Open Models from Review Process

White House limits AI review to closed systems

The Trump administration has designed its voluntary framework for testing advanced AI capabilities to apply exclusively to closed-source models, according to sources familiar with internal discussions. Open-source AI systems will face no restrictions under the framework, which the White House does not plan to make public.

The framework defines covered models as closed-source systems with state-of-the-art capabilities that pose potential national security risks. However, the criteria for what constitutes "state-of-the-art" or a genuine security threat remain undefined, creating uncertainty about which systems will fall under government review.

Why it matters

This approach creates a two-tier regulatory environment where closed commercial AI systems face government scrutiny while open models operate without oversight, despite ongoing debates about whether open-source AI poses greater proliferation risks. The lack of public transparency also means most companies and researchers cannot prepare for or understand the standards their systems might face.

30-day review with strict access controls

Under the framework reviewed during Tuesday meetings with select industry representatives, covered AI models would undergo a 30-day pre-release government review period. During this window, company employees would have limited access to their own systems, which must be stored in high-security environments with detailed access logs.

The review process will involve multiple administration officials rather than being centralized in a single agency. The framework explicitly states that nothing in its provisions should restrict open models after their release.

White House staff-level meetings on Tuesday included only certain companies, leaving others without knowledge of the framework's contents. It remains unclear which "trusted partners" will receive early access to advanced models under the arrangement, including whether any foreign governments qualify for this designation.

Narrow scope raises questions

Companies were encouraged during Tuesday's discussions to share models that are close to public release rather than early-stage systems, according to people briefed on the meetings. This guidance suggests the framework targets imminent deployments rather than research-phase development.

Many companies developing less advanced AI systems appear likely to fall outside the framework's scope entirely. The Trump administration's executive order specifies that the benchmarking process for assessing advanced cyber capabilities will be classified, and contains no requirement to publicly release the voluntary framework itself.

The details of the framework and Tuesday's meetings were first reported by Axios.