A search feature on the US Federal Register website, powered by Alibaba's Qwen model, was quietly removed around Wednesday. According to screenshots and web-archive snapshots reviewed by Reuters, the removal happened almost in step with the first social media posts pointing it out.
The Federal Register is run by the National Archives and publishes a daily list of newly proposed federal regulations. The feature that disappeared let users browse public comments on those regulations using natural-language queries. The site had offered several search options, and Qwen powered one of them.
The timing is what stings
On its own, the feature isn't much of a controversy. Qwen is an open-weight model, meaning its key components are publicly downloadable and can be modified as needed, and it typically costs less to run than closed models from Anthropic or OpenAI. Some experts note that the Federal Register's content is already public information, so using Qwen to search it doesn't appear to pose any immediate cybersecurity risk.
What stands out more is the timeline. Just last week, the FBI accused Alibaba of copying Anthropic's model technology to build its own AI tools, and next week the leaders of the United States and China are set to meet. A federal government website found running a Chinese model at this exact moment — and pulled that quickly — is enough to make the story travel.
That's also where caution is needed. When the feature was deployed remains unclear, and there's no official word on who made the call to remove it or why; both the National Archives and the White House have declined to comment. Reading "quietly pulled" as "the government scrambling to put out a fire" goes beyond what's actually been verified. Whether any data ever left the government's security perimeter is likewise still an open question, not a confirmed fact.
A choice now out in the open
This small episode is worth noting because it puts a bigger question in concrete terms: open-weight Chinese models are capable and cheap, which naturally appeals to budget-constrained agencies — but as long as a model comes from a Chinese vendor, buyers have to carry an extra layer of political and compliance risk under the current US-China climate.
For Chinese vendors, this is a real signal about what globalization looks like. Qwen supports many languages, is open-weight, and has racked up huge download numbers — technically, it has genuinely found its way into overseas government and enterprise use cases. But the moment a model steps into government procurement or critical infrastructure, there's a bar beyond how well it performs: a bar about where it comes from, and that bar isn't set by technical benchmarks. The Federal Register's approach this time — no explanation, just removal — says on its own that, in settings like this, avoiding controversy outranks functionality.
What's worth watching next is whether other US federal agencies start reviewing the Chinese models they have in use, and whether the sensitivity around model origin eases or tightens after next week's US-China meeting.
Sources: Reuters, CocoLoop, BusinessWorld; the removal timing, the FBI allegation, and the lack of an official response were verified against Reuters' screenshots and web-archive snapshots.