Report urges prohibitions in government systems
Two US lawmakers, House Homeland Security Committee Chairman Andrew Garbarino and House Select Committee on China Chairman John Moolenaar, both Republicans, on Friday urged a ban on Chinese-developed artificial intelligence models in US federal systems and high-risk critical infrastructure, warning that their growing adoption poses national security and cybersecurity risks. The two released an investigative report outlining 12 policy recommendations meant to reduce American dependence on Chinese AI technology and strengthen domestic alternatives.
The report urged the federal government to prohibit Chinese-developed AI models in government systems and federal contract work, remove them from sensitive functions within critical infrastructure, and prevent federal cybersecurity grants from being used to purchase them. It also recommended expanding existing federal restrictions beyond DeepSeek, the Chinese AI company whose technology has already been excluded from federal civilian agency systems under a Department of Homeland Security order.
“The investigation uncovered risks that American businesses, policymakers, and national security officials can no longer afford to overlook or downplay,” the lawmakers said in the report’s foreword.
Open-weight models and alleged capability extraction
The lawmakers warned that US companies, from tech startups to major corporations, are increasingly adopting Chinese-developed open-weight AI models because they offer competitive capabilities at significantly lower costs. Open-weight models allow users to download and modify the underlying parameters of AI systems, making them more customizable and potentially less expensive to operate than proprietary alternatives. The report found that Chinese models have become widespread in the American economy, although the extent of their use remains difficult to track, and it concluded that the US lacks sufficiently competitive domestic open-weight alternatives matching the price and performance of Chinese developers.
The lawmakers argued that Beijing is using government support, low prices and widespread distribution of AI models to expand its technological influence and create greater international dependence on Chinese systems. They also alleged that some Chinese developers have accelerated their progress by extracting capabilities from leading American AI systems through unauthorized techniques known as model distillation – training an AI model using outputs generated by another model, which allows developers to replicate certain capabilities without independently developing them from scratch. Federal investigations have identified instances in which Chinese developers used proxy networks and fraudulent accounts to collect outputs from American AI systems and reproduce their capabilities, according to the report.
The lawmakers also raised concerns about American data-labeling companies supplying training data to Chinese AI developers. The report warned that the risks could worsen as Chinese AI models grow more capable and acquire functions that conventional security evaluations may miss.
US companies examined in the investigation
The seven-month congressional investigation, launched in April, examined the adoption of Chinese AI models by several American technology companies, including Airbnb, Anysphere and DoorDash. Airbnb said it has incorporated AI models developed by Chinese technology company Alibaba into its customer support operations, while Anysphere, the developer of the Cursor AI coding platform, built its flagship proprietary model using a base model developed by China’s Moonshot AI. DoorDash reported that several families of Chinese-developed AI models were authorized and deployed within its operations.
Both committees said they had engaged with the three companies to examine the technical, commercial and cybersecurity considerations influencing their adoption of Chinese models. The investigation included document requests, company briefings, written responses, classified briefings and consultations with federal agencies, cybersecurity specialists, AI developers and academic researchers. The report did not establish that the use of Chinese-developed models by the companies had resulted in a cyberattack or unauthorized disclosure of sensitive information.
Recommendations to strengthen US AI capabilities
Beyond restrictions on Chinese models, the lawmakers called for a national initiative to accelerate the development of American open-weight AI systems and make trusted US technology more accessible internationally. They proposed establishing standards to identify the origins of AI models and creating an AI bill of materials that would allow companies and government agencies to determine which models and components are integrated into their systems.
Other recommendations included strengthening cybersecurity information-sharing arrangements, creating a dedicated channel for intelligence on AI-related threats and improving the federal government’s ability to assess vulnerabilities in foreign-developed models. The report also urged empowering the Cybersecurity and Infrastructure Security Agency (CISA) to coordinate reporting of AI-related cyber incidents and establishing an independent board to investigate significant cases. The lawmakers urged tougher action against unauthorized extraction of American AI capabilities and closer examination of the sale of US training data to Chinese laboratories, while recommending common standards with international allies and avoiding excessive regulation that could hinder American AI innovation.
The findings build on warnings issued in September by the National Security Agency, CISA and the FBI about large-scale efforts by China-based AI companies to extract capabilities from leading American models. The three agencies warned that the practices could have national security implications, particularly as advanced AI tools increase the speed and scale of cyber intrusions. The congressional committees said their investigation was intended to help lawmakers, federal agencies and businesses make informed decisions about protecting sensitive systems and critical infrastructure.
The proposed restrictions and other recommendations have not yet been enacted into law, and the committees said they would continue to oversee the issue.



You have reported this comment as spam. Your request is being reviewed.