Elements within the U.S. government have been weighing measures that could curb access to advanced Chinese open-source AI models, a shift that could indirectly strengthen the market positions of OpenAI and Anthropic.
Why this matters
Sources familiar with the discussions say past efforts to effectively block foreign open-source models resurfaced after the recent rise of the Chinese model Kimi. Pro-competition observers warn that limiting access to these cheaper, widely used Chinese models could reduce options for U.S. companies that rely on them.
What was discussed behind the scenes
- The U.S. Department of Commerce considered last year adding several Chinese AI laboratories to the Entity List, which would require licenses for U.S. interactions and substantially restrict access, a source close to the administration told Axios.
- The National Security Agency and the White House Office of the National Cyber Director weighed issuing an advisory last year about threats from Chinese AI labs, a move that would discourage U.S. firms from using their technology.
- The White House examined an executive order option that would allow U.S. companies to host Chinese models only if they could guarantee security and accept liability for breaches.
- Commerce circulated draft rules internally proposing to use its authority over domestic supply-chain security to target Chinese open-source models.
Why those steps stalled and why they returned
Officials worried that heavy-handed rules could stifle innovation, and those concerns initially blocked the proposals. Since then, personnel shifts — including the departure of former White House adviser Sriram Krishnan — and louder national-security voices, combined with more powerful Chinese AI offerings and renewed cybersecurity anxieties, have revived momentum for restricting Chinese models.
Public messaging and softer tools
Neither the White House nor the Commerce Department responded to requests for comment. Sources say the administration might not need an outright ban to push U.S. firms away from Chinese platforms. Instead, a combination of procurement rules, the threat of Entity List placement, and public-pressure campaigns could achieve that result. Another approach under consideration is publicizing potential backdoors and security weaknesses in Chinese models, stressing governance concerns and encouraging development of a more competitive U.S. open-source ecosystem.
The practical problem for restriction advocates
Even if policy measures make Chinese models harder to use, U.S. alternatives remain limited. Chinese models are often cheaper and more practical, which explains why many U.S. companies have adopted them.
Voices from the field
David Sacks, an outside White House AI adviser, wrote on X that AI policy is at a critical inflection point. He argued the leading closed labs—already a duopoly by AI model revenue—are pushing the government to eliminate open-source competition. Sacks has warned against regulatory capture that would favor the largest U.S. labs and hinder competition.
Sources close to the administration say the considered measures, even if not full bans, would have a chilling effect on Chinese open-source technologies and the American firms that depend on them. They also described frequent approaches by leading AI labs or their allies to the administration with ideas to curb open-source models every three to five months.
What to watch next
Chinese leader Xi Jinping appears emboldened in the AI race, and pressure is rising for the U.S. industry—open-source or proprietary—to respond.
Bottom line
OpenAI and Anthropic have long advocated safety regimes that, in some proposals, include licensing. While previously these proposals were mostly abstract, recent developments and the renewed policy push make the issue more immediate. As David Sacks put it, the major players have shown their cards; now other Silicon Valley actors who value open competition may need to state their positions clearly.
