US Federal Register website pulled Alibaba’s Qwen AI search tool after FBI named Alibaba a security risk
The National Archives removed an Alibaba-built Qwen AI search feature from the Federal Register website on Wednesday, September 17, 2026, after social media users noticed a US government site was running a Chinese model the FBI had publicly warned against.
Visitors to the Federal Register had been offered the option to use a small Qwen model to search public comments on proposed regulations. An archived copy of the site’s source code confirmed the model was taken down on Wednesday, meaning the service had been available for at least a day before it was pulled.
How a Chinese AI model ended up on a US government site
The Qwen search option was first spotted by an X user with the handle “tleilax___”, who posted a screenshot of the Federal Register site on September 15 showing the AI search tool in use. The discovery drew attention because the Federal Bureau of Investigation had, earlier in September 2026, named Alibaba among six leading Chinese firms it accused of conducting “industrial-scale distillation,” a practice the FBI says is helping China copy US frontier models and cut development time in the global AI race.
It is still unclear when the National Archives, the independent agency that runs the Federal Register, began offering the Qwen-powered search. The agency has not publicly commented on the removal, and the White House and the FBI also declined to comment.
What model was actually running
A Chinese-language report on Sina.com identified the specific model as a small open-weight “Qwen3 0.6B level” release, not Alibaba’s largest flagship. The model is open source, designed only to retrieve documents rather than reason over them, and was configured in a way that did not send government data to Alibaba or any third party.
For US agencies, models like this give the government more control over data because they can be downloaded and run locally instead of routing queries to larger external models. Sina.com also suggested these models stop counting as Chinese services once an American institution takes control of them.
Experts said the security risk was probably low
Georgetown University Law Professor Anupam Chander told Reuters that any actual risk depends on how the model was trained, and on whether any US data was processed by “Alibaba-controlled systems,” a point echoed by Senator Mark Warner (D-Va.). Chander added that the Federal Register’s content is already public, so the model was not working with sensitive government information.
Daniel Castro, president of the Information Technology and Innovation Foundation, called it an “insane” disconnect for a US agency to use an Alibaba model at the same time the FBI is encouraging stakeholders to stick to American systems.
Congress and the open-source AI debate
The episode landed in the middle of a broader Washington argument over how strictly the US should treat Chinese-developed AI. Many business owners who rely on open-weight models have asked the Trump administration not to restrict access to Chinese releases, and China has urged the US to listen to its own business community or risk falling behind economically.
A March 2025 policy submission to the US-China Economic and Security Review Commission warned that the US may already be at a disadvantage because it focused on maintaining a lead in frontier AI while letting its open-source ecosystem lag. The researchers noted that US export controls are calibrated around frontier training and advanced chips, and do not address the small-model deployment cycle, which uses less compute, draws on openly available base models, and builds advantage through application rather than pre-training. If the models that matter most for industrial AI are small, specialized, and open, current US policy may be aimed at the wrong layer of the competition.
Companies like Nvidia and Meta have pledged to help the US close that gap, and researchers have warned that losing a broad user base to China could set back America’s ability to set the technical standards and norms that will govern AI development for years to come.
US Rep. John Moolenaar (R-Mich.), who chairs the House China Committee, has taken the hardest line, saying “no federal government entity should use a Chinese AI model.” “Doing so only makes the federal government more dependent on Chinese AI models, and that is not in the national interest,” Moolenaar said.
FAQ
Did the Federal Register website use a Chinese AI model?
Yes. The Federal Register, run by the National Archives, briefly offered visitors an AI search option built on an open-weight Alibaba Qwen model. The feature was removed on Wednesday after social media users flagged it.
Why did the FBI warn about Alibaba AI models?
Earlier in September 2026, the FBI named Alibaba among six leading Chinese firms it accused of conducting “industrial-scale distillation,” a practice the FBI says helps China copy US frontier models and cut development costs.
Did the Qwen search tool pose a national security risk?
Experts quoted by Reuters said the risk was likely low. The tool used a small open-weight Qwen3 0.6B-level model, served only to retrieve public comments, and could be run locally without sending government data to Alibaba or any third party.
Related coverage
SEOScanPro
SEOScanPro runs a full technical audit of a site and shows the measured result behind every check. Open the AI visibility report.