A congressional advisory panel wants Chinese open-weight AI models barred from new deployments in national security and critical infrastructure systems unless their safeguards pass independent evaluation.
The US-China Economic and Security Review Commission made the proposal one of 10 key recommendations in its 2026 annual report to Congress, released November 18. Open-weight models are AI models whose trained weights are released for download.
“Congress should require enforceable protections for sensitive applications before these models become embedded in critical systems and costly to replace,” the commission wrote.
Under the proposal, Congress would direct the Secretary of Commerce to identify sensitive uses within 180 days. Those include systems that control critical infrastructure, perform national security functions, or process bulk sensitive personal data.
New deployments of covered models in those uses would be prohibited without a time-limited authorization based on independently evaluated safeguards.
Federal agencies and operators in designated sensitive sectors would have to inventory existing deployments and carry out mitigation or replacement plans. The commission also called for independent security testing, reporting of significant security incidents, and authority for Commerce to restrict or prohibit designated models where risks cannot be mitigated.
Providers serving sensitive applications would need documented ability to swap out covered models. The legislation would carry exceptions for security research and authorized national security activities.
A second key recommendation would require federal agencies, and state and local bodies that receive federal funds or tax credits, toh disclose any use of AI models developed by companies headquartered or organized in China, including models fine-tuned on or derived from them, wherever they are hosted.
The commission warned that wide adoption before the risks are assessed could leave US critical infrastructure, government networks, and defense-related systems exposed to threats the United States is not prepared to detect.
The report also said China’s progress appears driven at least in part by distillation of more advanced US models. Chinese military-linked researchers have used that technique to build defense AI from leading US models for surveillance, cyber operations, and drone control, a Reuters investigation found in July.
The proposal’s use of “Super Intelligence” in the standards center’s name tracks the federal terminology change that led the Pentagon to rename its chief AI office.
The commission asked Congress to fund the program, grant Commerce information-gathering and enforcement powers with penalties for noncompliance, and require annual reports on deployment exposure, compliance and remaining dependencies.