US Lawmakers Push to Ban Chinese AI Models in Federal Systems
By Staff, Agencies
Two Republican US lawmakers on Friday called for prohibiting the use of Chinese-developed artificial intelligence models across federal government systems and sensitive critical infrastructure, citing potential national security and cybersecurity threats.
House Homeland Security Committee Chairman Andrew Garbarino and House Select Committee on China Chairman John Moolenaar unveiled an investigative report containing 12 policy recommendations aimed at reducing US reliance on Chinese AI technologies and strengthening domestic alternatives.
The report called for banning Chinese-developed AI models from federal government operations and government contract work, removing them from sensitive critical infrastructure functions and preventing federal cybersecurity funding from being used to acquire such technologies.
It also recommended extending existing federal restrictions beyond DeepSeek, a Chinese AI developer whose technology has already been barred from federal civilian agency systems under a Department of Homeland Security directive.
In the report’s foreword, the lawmakers said their investigation had identified security concerns that American businesses, policymakers and national security officials could no longer afford to ignore.
They warned that companies across the United States, ranging from emerging technology firms to major corporations, were increasingly turning to Chinese open-weight AI models because of their competitive performance and comparatively low costs.
Unlike proprietary AI systems, open-weight models allow users to access, download and modify their underlying parameters, providing greater flexibility and potentially reducing operating expenses.
According to the report, Chinese-developed models have gained considerable traction across the US economy, although their overall adoption remains difficult to measure. It also found that American developers have yet to produce sufficiently competitive open-weight alternatives matching the affordability and performance of Chinese products.
The lawmakers argued that Beijing was leveraging government support, lower prices and broad international distribution to expand China’s technological influence and increase global dependence on its AI systems.
The report also accused some Chinese AI developers of obtaining capabilities from leading American models through unauthorized model distillation. The technique involves training one AI system using responses generated by another, allowing developers to reproduce certain capabilities without independently developing them from the beginning.
Citing federal investigations, the report said some Chinese developers had allegedly used proxy networks and fraudulent accounts to obtain outputs from American AI systems and replicate their capabilities.
It also raised concerns about US-based data-labeling companies providing training materials to Chinese AI developers.
The lawmakers warned that potential security vulnerabilities could become more significant as Chinese models become increasingly sophisticated and incorporate capabilities that conventional cybersecurity assessments might fail to identify.
The congressional inquiry, which began in April and lasted seven months, examined the use of Chinese AI technology by several American companies, including Airbnb, Anysphere and DoorDash.
According to the report, Airbnb has integrated AI models developed by China’s Alibaba into its customer support operations.
Anysphere, the company behind the Cursor AI coding platform, developed its flagship proprietary model using a foundational model created by Chinese AI firm Moonshot AI.
DoorDash also informed investigators that several families of Chinese-developed AI models had been approved and deployed within its operations.
The two congressional committees said they had consulted the three companies to assess the technical, commercial and cybersecurity factors behind their decisions to adopt Chinese AI technologies.
The investigation involved requests for documents, company briefings, written submissions, classified discussions and consultations with federal agencies, cybersecurity experts, AI developers and academic researchers.
However, the report did not establish that the companies’ use of Chinese-developed AI models had resulted in cyberattacks or the unauthorized exposure of sensitive information.
Beyond restricting Chinese AI products, the lawmakers proposed a national initiative to accelerate the development of American open-weight models and expand the international availability of trusted US-developed technologies.
Their recommendations also included establishing standards to verify the origins of AI models and introducing an AI bill of materials, allowing businesses and government agencies to identify the models and components incorporated into their systems.
Additional proposals focused on improving cybersecurity information-sharing arrangements, establishing a dedicated intelligence channel for AI-related threats and strengthening federal capabilities to evaluate security weaknesses in foreign-developed models.
The report further called for expanding the role of the Cybersecurity and Infrastructure Security Agency (CISA) in coordinating reports of AI-related cyber incidents and creating an independent board to investigate major cases.
The lawmakers also urged stronger measures against unauthorized extraction of American AI capabilities and closer scrutiny of US training data being sold to Chinese research laboratories.
They recommended coordinating common security standards with international allies while avoiding excessive regulation that could undermine American innovation in artificial intelligence.
The findings follow warnings issued in September by the National Security Agency, CISA and the FBI regarding large-scale efforts by China-based AI companies to obtain capabilities from leading American models.
The agencies cautioned that such practices could have national security implications, particularly as increasingly advanced AI technologies expand the speed and scale of potential cyber intrusions.
The congressional committees said their investigation was intended to provide lawmakers, government agencies and private companies with information needed to protect sensitive systems and critical infrastructure.
The proposed restrictions and policy recommendations have not yet become law, and the committees said they would continue monitoring the issue.
