Federal Register removes Qwen search tool as scrutiny emerges
The National Archives-run site briefly offered Alibaba’s Qwen to search public comments, raising questions after a federal AI advisory.
By James Whitfield · Staff Writer
3 min read
The Federal Register Qwen search tool was removed on September 17 after the National Archives-run website briefly offered Alibaba’s AI model to help visitors browse public comments on proposed regulations. Reuters reported that the episode drew attention because federal agencies had days earlier alleged Alibaba engaged in malicious, industrial-scale knowledge distillation involving U.S. frontier models.
Reuters said screenshots and an archived version of the site’s code showed the Qwen option was taken down on Wednesday, around the time posts about it appeared on social media. Reuters could not determine when the feature was introduced; the National Archives and White House did not respond to its requests for comment, while the FBI declined to comment. No official explanation for the removal was reported.
What was the Federal Register Qwen search tool?
Reuters described Qwen as an open-weight AI model from Alibaba. That means key parts of the system are publicly available, allowing developers to download and adapt the model for a particular task, unlike a service that necessarily sends prompts to a provider’s hosted system.
A September 8 joint advisory from the FBI, CISA and NSA named Alibaba and five other China-based companies. It alleged they had used aggressive, malicious and targeted knowledge-distillation campaigns at industrial scale to extract capabilities from U.S. AI models. The advisory said Alibaba used such activity to improve its Qwen family, while also noting that distillation itself can be a legitimate AI research technique.
Did the use of Qwen create a security risk?
Reuters reported that Georgetown Law professor Anupam Chander saw no apparent immediate cybersecurity risk from the Federal Register use, subject to how the model was trained. He said the site’s material was already public and therefore did not involve sensitive government information.
Senator Mark Warner, the vice chairman of the Senate Intelligence Committee, said the risk would depend on whether U.S. data crossed the government’s security boundary and was handled by Alibaba-controlled systems. Reuters did not establish the deployment’s technical or data-handling arrangement.
The allegations about Alibaba’s model-development practices do not by themselves show that the short-lived search feature was malicious or that it caused a data breach. A Chinese Embassy spokesperson in Washington called the copying allegations unfounded, Reuters reported.
The Office of the Federal Register provides access to federal laws, presidential documents, administrative regulations and notices, according to the National Archives. FederalRegister.gov says its XML-based site is an unofficial informational edition and directs people relying on documents for legal research to verify them against an official edition.
Representative John Moolenaar, the Republican chair of the House China Committee, told Reuters that federal entities should not use Chinese AI models, arguing that such use could increase government dependence on Chinese technology. The dispute therefore centers on both data security and the government’s broader approach to Chinese-origin AI models.
This story draws on original reporting from Ars Technica.