The National Archives quietly pulled an Alibaba AI search tool from the Federal Register website this Wednesday. The move followed a public outcry after users spotted the Chinese Qwen model powering searches for public comments on proposed regulations. This deployment created a stark contradiction: while the FBI recently accused Alibaba of “industrial-scale distillation” of US technology, a federal agency was actively using its software. The service was live for at least one day, confirmed by archived source code and screenshots shared on social media.
The timing could not have been worse for the agency. Earlier this month, the FBI named Alibaba among six Chinese firms allegedly stealing US frontier models to cut development costs. Daniel Castro, president of the Information Technology and Innovation Foundation, called the situation an “insane” disconnect. He pointed out the absurdity of a US agency relying on a model that top law enforcement officials claim is built on stolen intellectual property. It highlights a significant gap between national security rhetoric and operational reality within the government.
Security experts remain divided on the actual risk posed by the brief deployment. Georgetown University Law Professor Anupam Chander noted that the danger depends largely on how the model was trained. Senator Mark Warner added another layer of concern, questioning whether any US data was processed by “Alibaba-controlled systems” during that window. If user queries were sent to servers in China, the implications go beyond mere hypocrisy. However, if the model ran locally or on isolated infrastructure, the threat might be minimal.
Neither the White House nor the FBI has commented on the removal. The National Archives also failed to respond to requests for explanation. This silence leaves open questions about how such a vendor was selected in the first place. Did procurement officers miss the FBI’s warning? Or was this a low-stakes experiment that slipped through the cracks? Regardless of intent, the incident exposes the messy reality of integrating global AI tools into sensitive government workflows. Trust is hard to build when policy and practice collide this visibly.
