The National Science Foundation has notified the research community of a forthcoming policy that would prohibit the scientists it funds from collaborating with organizations on US government restricted-party lists, a move that would formalize the decoupling of American and Chinese research in AI and other strategically sensitive fields.

The stakes are unusually high for artificial intelligence. NSF is a primary funder of fundamental AI research at US universities, underwriting roughly $500 million across 27 National AI Research Institutes, the largest AI research program the federal government runs, and AI is one of the disciplines where American and Chinese researchers have co-authored most heavily. That overlap has also drawn the most national-security scrutiny. Research by the risk-analysis firm Strategy Risks and the Human Rights Foundation found that the Shanghai Artificial Intelligence Research Institute had co-authored dozens of papers with US institutions including MIT, Stanford, and UC Berkeley, and had acknowledged NSF funding in 13 of them, exactly the kind of tie the new policy is designed to cut.

The policy, issued as a “Dear Colleague Letter” providing advance notice rather than as a final rule, would bar three things on NSF-funded projects: using NSF funds to conduct research in collaboration with a restricted entity; senior or key personnel holding an appointment or position at one; and those personnel receiving research support from, or collaborating with, a restricted entity or its employees on their NSF-funded work. NSF said it had concluded that risk-mitigation measures for such projects were “not sufficient.”

Notably, the policy does not name China. It points instead to seven existing federal restricted-party lists, maintained by the Departments of War, Commerce, Treasury, State, and Homeland Security, the FCC, and Customs and Border Protection, which together name hundreds of Chinese universities and national laboratories. The effect falls overwhelmingly on China without a country being specified. The change is not yet final: NSF is taking public comments and intends to implement it in fiscal year 2027. It follows similar restrictions from the Department of Defense, NIH, and NASA, and sustained pressure from congressional Republicans on the House Select Committee on the CCP.

The result would push the AI field toward the separation that export controls and model-access limits have already imposed on chips and deployed models, extending the US-China divide from hardware and products back into the basic research that produces the next generation of both.

Sources: Science, NSF, Human Rights Foundation

–
By the Control Plane Editorial Team