From Ars Technica · · 2m
US agency caught using Chinese AI the FBI called malicious
The National Archives deployed Alibaba's Qwen model on a federal website while the FBI was publicly warning against exactly that.
In brief
The National Archives deployed Alibaba's Qwen model on a federal website while the FBI was publicly warning against exactly that. The incident exposes a disconnect between US national security rhetoric on Chinese AI and practical government adoption of open-source models that offer cost and control benefits…
Federal website quietly used Alibaba AI
The National Archives deployed Alibaba's Qwen search tool on the Federal Register website, allowing visitors to search public comments on proposed regulations using the Chinese model.
FBI had just called Alibaba malicious
Days before the discovery, the FBI named Alibaba among six Chinese firms allegedly conducting industrial-scale copying of US AI models to cut development costs and compete globally.
Social media users spotted the contradiction
A user on X posted a screenshot on September 15th showing the Qwen search option. An archived version of the site's code confirmed the model was removed the following Wednesday after the post went viral.
Experts say security risk was minimal
The Federal Register used a small, open-source Qwen model that only retrieved already-public documents. The model could run locally on government servers, avoiding data transfers to Alibaba.
“The site's content is already public, so the model was not working with sensitive government information.”
Why agencies preferred the Chinese model
Open-source Chinese models offer lower, predictable costs and give agencies more control by running locally rather than sending queries to external servers. This made them a default choice for many government uses.
US may be losing the AI standards race
Congressional research suggests the US focused too heavily on frontier AI while China built a dominant open-source ecosystem. If small, specialized models matter most, current US policy may be targeting the wrong competition layer.
Lawmakers demand a hard line
Rep. John Moolenaar, who chairs the House China Committee, said no federal agency should use Chinese AI models, arguing it increases government dependence on Chinese technology.
“No federal government entity should use a Chinese AI model.”
Business community pushes back
Many companies relying on open-source models have urged the Trump administration not to restrict Chinese AI access, warning that restrictions could harm US economic competitiveness.
In short
The incident exposes a disconnect between US national security rhetoric on Chinese AI and practical government adoption of open-source models that offer cost and control benefits.
