US government website used Chinese model the FBI called "malicious"

Officials removed an Alibaba Qwen search option from the Federal Register website after social media users highlighted its presence. A widely shared screenshot posted on X on Sept. 15 showed the Qwen choice; Reuters confirmed via an archived copy of the site’s source code that the option was taken down on Wednesday, indicating the tool was live for at least a day. The National Archives has not publicly commented on the removal, and the White House and FBI did not respond to requests for comment. The deployment attracted attention because the FBI recently accused Alibaba of participating in large-scale efforts to distill U.S. frontier AI models, listing the company among six Chinese firms in that report. Observers pointed to an inconsistency between that enforcement posture and the apparent use of an Alibaba-derived model on a U.S. government site. Daniel Castro of the Information Technology and Innovation Foundation called the situation an “insane” disconnect between policy messaging and practice. Security experts cited by Reuters said the specific Qwen instance on the Federal Register was unlikely to have created substantial risks. Georgetown law professor Anupam Chander noted that the site’s materials are already public, limiting exposure of sensitive government information. Reporting in China’s Sina.com described the deployed model as a compact, open-weight 'Qwen3 0.6B level' used solely for document retrieval and not for complex reasoning or forwarding government data to third parties. Defenders of using such open models argue they can increase government control over data when run locally, since they avoid sending queries to externally hosted large models. That local-deployment capability and lower, more predictable costs have made smaller open-weight retriever models a common choice for search tasks. But the episode feeds into a wider U.S. debate about whether Chinese-origin models — even when downloaded and operated domestically — should be treated with the same caution as externally hosted Chinese services. Legislators and some policymakers, including Rep. John Moolenaar who chairs the House China Committee, have called for federal agencies to avoid using Chinese AI models entirely.