U.S. Federal Register Site Used Alibaba Qwen for AI Search, Then Removed It

Media reported on 17 September 2026 that the U.S. Federal Register website had used Alibaba's Qwen model for AI search of federal rules and public comments. The search feature was taken down on 16 September 2026 after the arrangement drew attention on social media.

Qwen search removed after public attention

Media reports on 17 September 2026 said the U.S. government Federal Register website had been using Alibaba's Qwen artificial intelligence model to provide search over federal regulations and public comments. After the arrangement drew attention on social media, the search function was taken down on 16 September 2026 local time.

The Federal Register is operated by the National Archives and Records Administration (NARA). It publishes, each day, newly proposed U.S. federal rules and related information. According to the reports, as of 16 September the site offered several search methods, including an AI search tool based on Qwen. The tool was used mainly to help users browse public comments submitted on proposed federal rules. It was not clear when the tool had first been deployed.

Takedown timing

Using website screenshots and archived source code, reporters confirmed that the Qwen-based search function was removed on 16 September. The removal coincided with the period when social media users began discussing a U.S. government website that used a Chinese AI model. NARA and the White House did not respond to requests for comment from Reuters. The Federal Bureau of Investigation (FBI) declined to comment. As of the time of the reports, the U.S. government had not explained why the Federal Register site chose Qwen, how the model was deployed, or why the search function was withdrawn.

Expert views on data exposure

The episode drew attention in the context of recent U.S. government allegations of technical "distillation" involving Chinese AI companies. AI and cybersecurity specialists interviewed by the press said use of Qwen on the Federal Register site did not appear to create a direct national security risk by itself. Anupam Chander, a professor at Georgetown University Law Center, said the Federal Register handles information that is already public, so Qwen did not come into contact with sensitive government information. He added that residual risk still depends on how the model is trained and deployed.

Senator Mark Warner, vice chair of the Senate Intelligence Committee, said one key question is whether U.S. government data left the government's own security boundary and entered systems controlled by Alibaba for processing. Security specialists noted that local deployment can give an agency more direct control over data and can reduce cases in which sensitive information must be sent to an external AI provider.

Policy contrast

The dispute centered on a gap between U.S. government statements on Chinese AI technology and the actual use of a Chinese-developed model on a government site. Daniel Castro, president of the Information Technology and Innovation Foundation (ITIF), said that while the United States and China compete intensely over AI models, a U.S. government agency had chosen a model developed in China.

The episode also came as U.S. debate over AI safety and regulation continued. The U.S. government has recently intensified technical-security and intellectual-property allegations against Chinese AI companies. At the same time, Chinese open-weight models have been adopted by users that include U.S. institutions and companies. In April 2026, a relevant U.S. House committee sent a letter to Airbnb asking it to explain its use of Qwen and to address possible national security risks from integrating Chinese AI tools into platforms used by U.S. customers.