Key facts
- The US government's Federal Register website used Alibaba's Qwen AI model for user searches.
- The FBI accused Alibaba of "malicious" copying of US rival Anthropic's technology.
- The Qwen search tool was removed from the website on Wednesday.
- The Federal Register website is run by the National Archives.
- The Qwen model is an open-weight model, allowing modification by developers.
- The content searched on the Federal Register website is public.
The US government's Federal Register website, which lists proposed federal regulations, briefly employed Alibaba's Qwen AI model for user searches. This occurred despite the FBI's recent accusation that Alibaba engaged in "malicious" copying of technology from US AI company Anthropic. The Qwen-powered search option was available on Wednesday and was removed around the same time social media posts began highlighting its use, according to Reuters' review of screenshots and website source code.
US officials have stated that Chinese AI companies are copying US technology on an "industrial scale." The FBI declined to comment on Thursday, and the National Archives and the White House did not respond to requests for comment.
Experts suggested that the use of the Chinese AI tool on the Federal Register's website, which deals with public information, may not have posed immediate national security risks. However, it raised questions about the federal government's stance on Chinese AI models, particularly given the ongoing competition between the US and China in AI development. Daniel Castro, president of the Information Technology and Innovation Foundation, called the situation "insane," highlighting a disconnect between US policy and agency actions.
Qwen is an open-weight model, meaning its key components are publicly available, allowing developers to modify it for various uses, often at a lower cost than closed models. Security experts note that open-weight models can be run on an organization's own infrastructure, offering greater control over sensitive data. Georgetown Law Professor Anupam Chander indicated that the risk depends on how the model was trained, but since the Federal Register's content is public, sensitive government information was not involved. Senate Democrat Mark Warner stated that the risk hinges on whether US data left the government's security boundary and was processed on Alibaba-controlled systems. Previously, US House committees had sent a letter to Airbnb, which also uses Qwen, inquiring about national security risks associated with Chinese AI tools.