US government website powered by Chinese model. FBI previously called it a threat

Government infrastructure on a hostile stack

According to findings by Ars Technica, the search system on the US government website Federal Register ran for some time on an open-source model from Chinese developers. The integration of this tool took place despite the fact that the FBI had previously identified technologies of the same origin as a security threat and warned of the risk of state actors inserting backdoors into open code.

Cheap open-source won over compliance

The incident shows the gap between declared geopolitical vigilance and the reality of government IT procurement. Administrators likely reached for the Chinese model because it was technically competent and freely available, without the code undergoing a thorough origin check within the supply chain. On paper, there's an arms race; in practice, what can be quickly set up for free from GitHub wins.

Risk of eavesdropping versus data poisoning

Deploying a Chinese model in search does not in itself represent an immediate leak of sensitive data, as the Federal Register contains public information. The fundamental risk, however, is the integrity of the system. If an attacker controls the search model, they can distort results, hide certain documents, or manipulate how citizens interpret American laws and regulations.

Origin, not just performance, will be audited

This misstep will lead to pressure to introduce much stricter rules for the use of open-source artificial intelligence in government administration. We can expect the creation of government blacklists for specific open-source repositories and models, which will make life harder for developers building infrastructure on globally available code.

Lilith's verdict

The bureaucratic apparatus may think it's in the middle of a tech Cold War, but the average state IT administrator would rather reach for banned Chinese code than build something from scratch.