白宫完成AI审查框架：闭源模型发布前需审查30天

The White House has finished a framework that gives the government 30 days to review the most advanced AI models before they ship.

None of it applies to open-weight AI, because the policy covers only closed-source products.

The Trump administration invited staff from OpenAI, Anthropic, Google, Meta, Nvidia, and other major AI firms to the White House on Tuesday to present its new AI oversight framework. AI developers can voluntarily submit new models to the federal government up to 30 days before public release.

The document reportedly says nothing inside it should be read as restricting freely downloadable weights after release.

This oversight framework stemmed from an executive order from June-2026.

The June-26 executive order called for a benchmarking process, run with the NSA and other agencies, to decide which products carry national security risk.

The framework built on that order defines a covered model as closed-source, state-of-the-art, and a security risk, without defining any of those three terms.

Once something qualifies, it sits in a high-security environment for up to 30 days, with logs recording every person who opens it.

Even inside government, access is restricted, and the review pulls officials from across the administration rather than one agency.

You will not get to read the original rule document, since there is no plan to publish the rules and the order does not require it. For now, the public remains in the dark.

Companies not in Tuesday's room still do not know what it says.

Those that were there were told to submit products close to launch rather than early builds, which writes smaller developers out of the process.

The trusted partners who get early access are undefined too, and nobody will say whether foreign governments qualify.

wsj. com/tech/ai/white-houses-ai-guidelines-exempt-u-s-open-models-from-government-review-74924eb8