National Archives Removed Alibaba AI Search Tool From Federal Register Site

The National Archives removed an Alibaba AI search option from the Federal Register website after users publicly noticed it, turning a short-lived feature for searching regulatory comments into a test of how the US government should treat Chinese-origin open models.

The site had offered visitors an AI-powered way to search public comments on proposed regulations using an Alibaba Qwen model. A screenshot posted on X on September 15 showed the option, and an archived version of the site’s source code confirmed that it was removed Wednesday. The service was therefore available for at least a day; its precise start date is unclear.

The model was identified as a small, open-weight Qwen3 0.6B-level model, rather than Alibaba’s flagship system. That distinction shapes the security case: experts told Reuters that the deployment was unlikely to create substantial risk because the material being searched was already public and the model could be downloaded and run locally, without sending government data to Alibaba or another third party.

The removal landed shortly after the FBI named Alibaba among six leading Chinese firms it accused of conducting industrial-scale distillation of US frontier models. The FBI allegation is separate from the Federal Register deployment, but it made the government’s use of an Alibaba model look sharply at odds with law-enforcement warnings about Chinese AI firms.

The episode does not establish that the Federal Register tool exposed sensitive data. The expert view described a limited search use case with public inputs and an option for local operation. But Moolenaar’s objection rests on a different concern: even a low-risk deployment can, in his view, make federal agencies more dependent on Chinese AI technology.

The National Archives had not commented on the removal, and neither had the White House or the FBI. That leaves unanswered whether the tool was withdrawn over a specific security assessment, a policy concern about its origin, or the public attention itself. For agencies weighing open models, the harder question is no longer only where data goes, but whether a model’s provenance should determine whether it can be used at all.