Trump Administration Plans 30-Day Government Review for Advanced AI Models; Framework Due Aug. 1

The Trump administration is advancing a regulatory framework for frontier AI models, requiring companies to submit their most advanced models to government review before public release. The White House National Cyber Director's office distributed a draft of the framework to OpenAI, Anthropic, and Google about two weeks ago, and the three companies subsequently submitted joint revision suggestions. The framework stems from an executive order signed by Trump in early June this year, with a completion deadline of August 1. The framework aims to establish a more standardized process for government review of powerful AI models, focusing on frontier models capable of rapidly identifying cybersecurity vulnerabilities.

The issuance of the executive order is directly linked to Anthropic's frontier model Mythos, which possesses offensive cybersecurity capabilities. Anthropic voluntarily delayed its large-scale public release. The Trump administration seeks to maintain a hands-off approach to AI regulation while ensuring effective oversight of national security risks from new models. The executive order requires AI labs to provide the federal government with an access window of up to 30 days before releasing covered frontier models to other partners, and gives government agencies 60 days to design specific frameworks.

One of the key disagreements over the framework is how to define the scope of frontier models. The criteria for judging frontier models include model size, deployment method, and specific-purpose capabilities measured by third-party evaluations.

The treatment of open-source models is another core issue. Parallel discussions are underway between companies and the government regarding whether open-source models should be granted voluntary review exemptions based on their capabilities. Those supporting exemptions argue that publicly available open-source models in the US currently lag behind their Chinese counterparts, and reducing regulation can ensure room for catch-up. If the framework does not classify open-source models as frontier models, these models will not be able to participate in the program and will be cut off from access to relevant resources.

The Trump administration insists that compliance with the framework is not mandatory, but its actual binding force is under scrutiny. The White House's previous imposition of export controls on Anthropic has raised doubts about the credibility of voluntary commitments. The White House has told some companies that the agencies responsible for implementing model reviews will be jointly undertaken by the National Security Agency and the AI Standards and Innovation Center under the Department of Commerce.