Will U.S. AI Export Controls Stall Global Large‑Model Development?

The U.S. government is preparing a new AI restriction line, stating its goal is to protect American artificial intelligence from countries such as China, Russia, North Korea and Iran, and is planning to impose export controls on the most advanced proprietary AI large models.

On May 9, a bipartisan group of lawmakers announced a technology‑restriction bill that would make it easier for the Biden administration to apply export controls to AI large models, aiming to safeguard U.S. technology from foreign malicious actors.

Large AI models, the core technology behind ChatGPT, DeepMind and similar systems, currently face no U.S. restrictions on domestic companies selling their closed‑source models worldwide.

To define AI model export controls, the United States may adopt the threshold from the October 2023 AI executive order, which is based on the computational power required to train a model; once that level is reached, software providers must report their AI development plans to the Commerce Department and submit test results.

The bill, introduced by Representatives Michael McCaul, John Molenaar, Raja Krishnamoorthi and Susan Wild, would explicitly grant the Commerce Department authority to prohibit U.S. persons from collaborating with foreigners on AI systems that pose national‑security risks.

Legislation also seeks to ensure future AI export regulations can withstand constitutional challenges and will be shaped according to input from the Biden administration.

The United States is increasingly concerned that competitors could use large AI models to process massive text and image data, generate content, launch aggressive cyber attacks, or even create powerful biochemical weapons.

To date, the Commerce Department and the White House have not responded to requests for comment on the bill.

Under current law, the Commerce Department can more easily restrict the export of open‑source AI models, but establishing appropriate management standards will be difficult; AI policy experts estimate China is only about two years behind the United States in AI software development.

If approved, the measure would remove export‑regulation obstacles for open‑source AI under the International Emergency Economic Powers Act and give the Commerce Department clear, definitive authority over AI systems.

Sources indicate China heavily relies on many Western open‑source large models, such as Meta’s Llama series; Chinese media cited the Beijing Institute of Artificial Intelligence saying most domestic AI models are built using Llama, and the bill would pose a critical challenge to China’s AI progress.

AI startup 01.AI, founded by former Google executive Dr. Kai‑Fu Lee, attracted attention last November when engineers discovered its Yii‑34B model was based on Meta’s Llama system, sparking strong industry reaction.

In addition to AI model restrictions, the Commerce Department has revoked export licenses for chip manufacturers Intel and Qualcomm supplying chips to Huawei, focusing on Huawei’s laptops and smartphones; the department stated it continuously assesses export‑control measures to best protect national security and diplomatic interests.

House Foreign Affairs Committee Chairman Michael McCaul emphasized that the measures would block China from developing advanced AI technology, and the United States is also considering sanctions against six Chinese firms suspected of supplying chips to Huawei, which has been on a trade blacklist since 2019.

China’s Ministry of Commerce spokesperson criticized the U.S. restrictions on civilian‑use chips and the cutting off of supplies to Chinese firms as economic coercion that violates WTO rules, harms U.S. business interests, and contradicts U.S. commitments not to decouple from China.