Technology
US government briefly deployed Chinese AI model despite FBI warnings about Beijing's copying practices
The National Archives removed Alibaba's Qwen search tool from the Federal Register after days of public use, creating an awkward optics problem as the FBI investigates Chinese firms for allegedly stealing US AI technology.
Ars Technica
The National Archives took down a Chinese artificial intelligence tool from its Federal Register website this week after the deployment became public, according to Reuters reporting. The move highlights a stark contradiction in how the US government is approaching the competition with Beijing over AI dominance.
The contradictory deployment#
For at least a day in mid-September, visitors to the Federal Register—the official government site for posting proposed regulations and public comments—could use Alibaba's Qwen AI model to search documents. The option became visible to observers on social media, who flagged the incongruity with recent federal law enforcement activity.
Just weeks earlier, the FBI had publicly named Alibaba as one of six major Chinese companies allegedly engaged in "industrial-scale distillation," a practice the agency contends violates intellectual property rights and allows Chinese rivals to accelerate their AI development at reduced cost.
Silent removal, no explanation#
As of Wednesday, the Qwen tool was no longer available. The National Archives—the independent agency responsible for the Federal Register—has not publicly explained when it added the feature, why it chose Alibaba's model, or the circumstances of its removal. Neither the White House nor the FBI has commented on the matter.
Daniel Castro, president of the Information Technology and Innovation Foundation, characterized the situation as fundamentally at odds with itself. He told Reuters it made no sense for a US government agency to rely on an Alibaba product while federal law enforcement was actively warning the private sector against such choices.
Security questions remain open#
Experts cautioned that any genuine security risk would hinge on technical details. Georgetown law professor Anupam Chander suggested that how the model was trained—and specifically whether it involved US proprietary techniques—would determine if harm occurred. Senator Mark Warner (D-Va.) added that the critical question was whether any American data had been processed on systems controlled by Alibaba.

