Federal Register searched comments with Alibaba's Qwen

A search box on a federal website was running on a model built in China, and it disappeared on Wednesday at about the same time people started posting about it, Reuters reports. The site is the Federal Register, run by the National Archives, and the model is Alibaba's Qwen, the same Qwen the FBI pointed at a week earlier when it accused Alibaba of copying Anthropic's models.

At a glance

• The Federal Register, which publishes the daily list of newly proposed regulations, offered a Qwen-powered tool for browsing public comments, sitting alongside the site's other search options on Wednesday.
• Qwen is open-weight, so developers can download it, modify it for a specific job and run it on their own machines, often at lower cost than closed models from Anthropic or OpenAI.
• Georgetown Law professor Anupam Chander said the use of Qwen did not appear to pose an immediate cybersecurity risk, since the site's content is already public, though it depends on how the model was trained.

If you have not been following the row: the FBI last week accused Alibaba of copying Anthropic's models to build its own AI tools. US officials said Chinese AI companies were engaged in "malicious" copying of US technology on an "industrial scale", days before a meeting between the two countries' leaders. A spokesperson for the Chinese embassy in Washington said the copying allegations were "unfounded".

The Qwen search was live on Wednesday and gone by the end of it

The Federal Register's job is mundane and useful: it lists newly proposed federal regulations every day and collects the public comments people file on them. The Qwen-powered tool was one way to browse those comments, offered next to the site's other search options.

It came down on Wednesday, around the same time social media posts about it appeared, according to screenshots and an archived version of the site's source code reviewed by Reuters. When the search was deployed was not established. The episode surfaced during a wider US debate over whether regulatory safeguards on AI are needed or would leave the country at a competitive disadvantage.

What does open-weight actually mean for the risk here?

It means the key components of the model are published, so anyone can pull them down and run them. Security experts say open-weight models can run on an organization's own IT infrastructure, which gives it more control over data that would otherwise be handed to an outside provider. The closest everyday comparison is the difference between sending your files to a contractor and buying the recipe so you can cook in your own kitchen.

That distinction is exactly where the disagreement sits. Chander said the Federal Register's content is already public, so the model was not touching sensitive government information. Senator Mark Warner, the Virginia Democrat who is vice chairman of the Senate Intelligence Committee, said in a statement that the risk turns on whether US data left the government's security boundary and was processed on Alibaba-controlled systems.

Moolenaar says no federal entity should use a Chinese AI model

Daniel Castro, president of the Information Technology and Innovation Foundation, told Reuters there was a disconnect between the fierce US-China race to build better models and a US agency picking a Chinese one. "It's one of the most insane things I've ever seen," he said.

Representative John Moolenaar, the Michigan Republican who chairs the House China Committee, said no federal government entity should use a Chinese AI model, because it only deepens dependence on them. Congress has been here before with a private company: in April the House committees on China and homeland security wrote to Airbnb, which also uses Qwen, asking for details as part of an investigation into the national security risks created by integrating Chinese AI tools into platforms Americans use.

The gap in the reporting is the operational one. Reuters could not establish when the Qwen search went live, which means Warner's test cannot be settled from what is public: whether queries ran inside government infrastructure or on Alibaba-controlled systems is simply not stated, and in our view that single fact matters more than the political noise around it. Even the reassurance comes with a string attached, since Chander tied his reading to how the model was trained, which open weights on their own do not reveal.

Whether the search comes back

The leaders of the two countries meet next week with the FBI's accusation still fresh, and Alibaba's denial standing through its embassy. On the smaller question, nothing has been said about whether the Federal Register restores a Qwen-powered comment search, swaps in a different model, or drops the feature. The April letter to Airbnb shows what a congressional follow-up looks like; whether the same committees send one to a federal agency is open.