US Government Website Used Chinese AI Model
A United States government website deployed an artificial intelligence model from a Chinese company for users seeking proposed federal regulations. This action occurred even though the Federal Bureau of Investigation recently accused the model's developer, Alibaba, of deliberately copying technology from its US rival, Anthropic.
The website for the Federal Register, overseen by the National Archives, offered the search tool using Alibaba's Qwen model. This facility was available recently alongside other search options. The exact date of its initial deployment remains unspecified. The search function powered by Qwen was subsequently removed around the time social media discussions about its presence commenced, according to screenshots and an archived version of the site's source code reviewed by Reuters.
The FBI, some time ago, formally accused Alibaba of replicating Anthropic's models to construct its own artificial intelligence tools. This development amplified tensions between Washington and Beijing, preceding a planned meeting between the leaders of both nations. United States officials, during that period, characterised Chinese AI firms as engaging in "malicious" copying of American technology on an "industrial scale." But the Chinese embassy spokesperson in Washington later stated these allegations of copying were "unfounded."
The Federal Bureau of Investigation did not offer further comment on the matter. Both the National Archives and the White House did not respond to queries for information. The employment of Qwen by the US government emerged amidst a global re-evaluation of AI safety protocols and an ongoing debate within the United States regarding the necessity of regulatory safeguards for artificial intelligence, and whether such measures might impede the nation's competitive standing.
Artificial intelligence specialists indicated that the presence of the Chinese AI tool on the Federal Register's website likely presented no immediate national security risks. However, its use raised concerns about potential inconsistencies with the federal government's stated positions on Chinese AI models. Daniel Castro, president of the Information Technology and Innovation Foundation, identified a clear divergence between the intense competition for superior AI models between the US and China, and a US agency's decision to implement a Chinese-made AI model. He described it as "one of the most insane things I've ever seen."
The Federal Register website, which circulates a daily compilation of newly proposed regulations, provided the Qwen model as a means for browsing public comments. Qwen operates as an open-weight model, meaning the core components of its system are publicly accessible. Devs can acquire and modify this model for specific applications, often at a reduced cost compared to proprietary closed models from entities such as Anthropic and OpenAI. And security experts suggest that open-weight models can be run within an organisation's existing IT infrastructure, granting greater oversight of sensitive data that might otherwise be shared with external providers.
United States Representative John Moolenaar, a Michigan Republican who chairs the House China Committee, issued a statement condemning the use of Chinese AI models by federal government entities. He asserted that such practices foster increased reliance on Chinese artificial intelligence models, which he deemed contrary to the national interest.
Georgetown Law Professor Anupam Chander observed that the deployment of Qwen did not appear to pose an immediate cybersecurity threat. This assessment, however, hinges on the specifics of the model's training. Given that the Federal Register website's content is already public, the model was not processing classified government information. Democratic US Senator Mark Warner of Virginia, vice chairman of the Senate Intelligence Committee, stated that any risk would be contingent on whether US data left the government's security confines to be processed on systems controlled by Alibaba.
Earlier this year, the US House committees on China and homeland security despatched a formal letter to Airbnb, another organisation utilising Qwen. This correspondence sought comprehensive details regarding its employment of the AI tool. So, this forms part of an ongoing inquiry into the potential "national security risks created by the integration" of Chinese artificial intelligence tools into platforms accessible to American citizens.
A US government website used Alibaba's Qwen AI model despite FBI accusations of technology theft from Anthropic.
The Qwen search tool was recently removed following social media attention.
Experts generally assessed no immediate cybersecurity risk, but noted the contradiction in US policy messaging.
Concerns persist over US dependence on Chinese AI models, as voiced by politicians.
Qwen is an open-weight model, allowing developers to modify and run it on their own IT infrastructure.
Source: Reuters
