English

PLUS ULTRAAlibabaAlibaba Cloud

US officials remove Alibaba-developed Qwen model from Federal Register website

PLUS ULTRA by Amenoyomi

US government officials removed a Chinese AI search tool from the Federal Register website on Wednesday, following observations from social media users that the site was using Alibaba’s Qwen model.

The change occurred after it was noted that the National Archives, which runs the Federal Register website, was providing visitors an option to use the Qwen model to search public comments on proposed regulations. This deployment comes after the FBI named Alibaba among six leading Chinese firms allegedly conducting "industrial-scale distillation," an agency claim that such practices help competitors reduce development time and costs.

While some experts noted that the use of a small, open-weight version of Qwen—reported by Sina.com as a Qwen3 0.6B level model—likely posed minimal risk because it was used for searching public documents and could be run locally, lawmakers expressed concern. Representative John Moolenaar, chair of the House China Committee, stated that no federal government entity should use Chinese AI models, arguing that doing so increases national security dependence on Chinese technology.

PLUS ULTRAby Amenoyomi

The perceived low risk of the Qwen deployment stems from the specific nature of the model used. As a small, open-weight model—identified as being at the Qwen3 0.6B level—it can be downloaded and executed locally within the agency's own infrastructure. This technical setup allows the government to maintain control over data by avoiding queries to external, Alibaba-controlled systems, which is why some experts argued it posed no substantial security risk when searching public documents.

Despite these technical safeguards, the FBI and some lawmakers view the issue as a strategic vulnerability rather than a data leak concern. The primary risk is the "locking in" of US government reliance on Chinese technology. From this perspective, the use of Chinese-developed models in federal operations creates a systemic dependency that could undermine national interests, regardless of whether the specific implementation is local or secure.

This tension highlights a gap in current US AI policy, which has focused heavily on restricting the high-end semiconductors needed for "Frontier AI" training. Such controls do not address the deployment cycle of small, specialized models that require far less computing power and rely on existing open weights. While the US has prioritized the lead in the largest models, China has moved to fill the ecosystem for smaller, industrial AI tools, potentially placing the US at a competitive disadvantage in the application layer.

Sources

  1. US government website used Chinese model the FBI called "malicious" (Ars Technica AI, 2026-09-18)