AI framework excludes open models from US model reviews

AI framework plans under White House review would exclude open models while creating a 30-day check for closed frontier systems.

People familiar with the matter said the voluntary process is being shaped inside the Trump administration and is not expected to be released publicly. The framework described to participants would apply to closed-source frontier models with leading-edge capabilities and potential national-security concerns.

The exclusion matters because open models can be downloaded, modified and redistributed after release, making them harder to govern through a pre-launch review. The plan, as described by people briefed on White House meetings, says it should not be read as a restriction on open models once they are already public.

Closed models draw review

The framework would create a government review window before certain advanced systems reach the market. During that 30-day period, access inside companies would be narrowed, and the models would be kept in high-security settings.

People briefed on the discussions said the plan also calls for detailed records showing who accessed the systems. The review would involve several administration officials rather than putting one office or agency in sole control.

The document’s threshold is still unclear. People familiar with the meetings said it does not define precisely what counts as a leading-edge capability or what level of risk would trigger national-security treatment.

Tuesday meetings left gaps

The White House held staff-level meetings with industry representatives on Tuesday to walk through the framework. Companies that were not included in those sessions do not yet have the same view of how the process would work, according to people familiar with the matter.

That uneven access could become a practical problem for developers planning product launches, safety testing and customer commitments. A voluntary process can still shape business behavior if companies believe early government review will affect market confidence or future policy treatment.

The plan also leaves open which “trusted partners” might see advanced models early. People familiar with the discussions said it is not clear whether foreign governments could qualify, a question with direct consequences for allied cooperation and export-sensitive technology.

Open models remain outside

The carveout for open models reflects a central split in AI policy. Closed systems can be held inside controlled infrastructure before launch, while open releases can spread quickly across labs, startups, researchers and individual developers.

For the White House, that distinction creates a narrower review path that is easier to administer but less comprehensive. For companies building closed frontier systems, it could mean more formal contact with officials before release and tighter internal controls during the review period.

For the wider AI sector, the incentive effects are less settled. If closed models face heavier pre-release scrutiny while open models remain outside the framework, some developers may see openness as a way to avoid review, while others may prefer closed systems to maintain commercial control and security discipline.

Three paths for AI policy

If the current structure holds, the macro effect would be a more predictable US review channel for a subset of powerful AI systems, while open-model governance remains fragmented. Companies with closed frontier models would need stronger access logs and secure environments; the broader industry would operate under a two-track regime.

If the administration later defines national-security risk more broadly, the process could cover more systems and slow some launches. That would give officials more visibility into advanced capabilities, but it could raise compliance costs for developers and push smaller firms to rely more heavily on open tools.

If early access for “trusted partners” includes foreign governments, the framework could become part of allied AI coordination. If it is limited to domestic actors, companies may face separate approval expectations across jurisdictions, and the industry could see a patchwork of model-review practices rather than a shared standard.

The immediate questions are concrete: how the White House defines a covered model, which officials get decision-making authority, who qualifies for early access and whether companies outside Tuesday’s meetings receive the same guidance. Until those points are clarified, the framework remains influential but incomplete.