The Donald Trump administration has decided to exclude open-source models from an upcoming review system designed to evaluate the performance and risks of advanced artificial intelligence (AI) models prior to their release, according to U.S. media reports.

Citing online outlet Axios and CNN on August 6 (local time), the AI review system being prepared by the White House is set to apply only to powerful "closed-source" models whose underlying source code is not publicly disclosed, such as Anthropic's Claude and OpenAI's ChatGPT.

In a draft presented to industry stakeholders on August 4, the White House defined the "frontier models" subject to review as closed-source models equipped with state-of-the-art performance that could pose risks to national security.

Axios reported that clear criteria were not provided regarding what level of performance constitutes "state-of-the-art" or which risks are classified as national security-related threats.

Additionally, open-source AI models were excluded from the new review framework.

Consequently, open-source AI models—such as DeepSeek, which allow users to download, modify, and adapt model weights and inference code—are expected to be exempt from U.S. government review.

AI companies subject to the review will undergo a 30-day evaluation by the U.S. government before releasing their models, during which access to the models by company employees may also be restricted.

Rather than a single government department or agency, officials from the White House and multiple departments within the administration are expected to participate in the review process.

The New York Times (NYT) noted that this measure formalizes the Trump administration's oversight of major AI companies like OpenAI and Anthropic, while pointing out that it "pushed aside urgent issues related to AI, including potential threats posed by increasingly powerful open-source models developed by Chinese firms."

(Photo: AP, Yonhap News)