A US military AI chatbot mistakenly classified the cargo of a Chinese freighter as material for a nuclear weapons program in the spring of 2026, CNN reported on September 18. Armed soldiers were already preparing a boarding operation, and fighter jets were already in the air when officers noticed the error at the last minute.
Two AI queries generate an official-looking report
According to CNN, an analyst from the Special Operations Command in the Pacific initially used the chatbot to cross-reference open sources with classified signals intelligence on a Chinese freighter. The system misread entries on the cargo manifest and assigned the cargo to components for a nuclear weapons program. In a second step, the analyst used the same chatbot to format the erroneous results into the style of a regular intelligence report. Neither the manifest nor the origin of the alleged components was verified by a second person before it was forwarded. This formally official document then spread through military command channels without anyone rechecking the underlying raw data. Senior commanders trusted the presentation of the report and approved a boarding action against the ship, because an AI-formatted report was hardly distinguishable from one written by a human. The command has reportedly been using chatbots on a trial basis for months to speed up the analysis of raw data; a binding approval process for their results was missing in this case, however.
Fighter jets are already in the air when the error surfaces
Armed US soldiers were preparing to board the ship, and combat aircraft were already taking off. Only a subsequent review of the intelligence data revealed that the chatbot had misidentified the cargo; the operation was aborted at the last minute. A person involved in the intelligence work described the original report to CNN as entirely false and said it “almost started a war” – it remains unverified what was actually on board the ship. A boarding action against a Chinese merchant vessel would have carried the risk of a direct military confrontation between the two nuclear powers, in the middle of the ongoing US war against Iran, which had begun on February 28, 2026. CNN does not name any publicly known consequences for the analyst or the commanders involved. The incident took place against the backdrop of a growing role for AI in military operations: just in July, the US Navy had presented its own AI strategy for warships, which aims to feed sensor data more directly into combat decisions.
Hegseth pushes AI use, critics call for oversight
Defense Secretary Pete Hegseth is said to be pushing the use of AI at the Pentagon especially hard. Critics see the near-miss as part of a broader pattern of insufficient human oversight of AI-assisted military decisions in the ongoing war. On the very first day of the war, February 28, 2026, a targeting system supported by the Maven Smart System software is said to have contributed to an airstrike on a school in Minab that killed more than 120 children; a Bloomberg investigation documents the case, and a UN investigation assessed the attack in mid-September as potentially unlawful. GovAI researcher and Army veteran Jake Steckler argues, according to TechCrunch, that service members need to understand the uncertainty inherent to large language models – the answer is not less AI use, but better oversight. Together, the two incidents are fueling a debate in Washington over how much trust commanders should place in automatically generated documents without verification.
What matters now is whether the Pentagon responds to the incident by introducing binding review steps for AI-assisted intelligence reports before commanders can use them for operational decisions. So far, the Department of Defense has not commented publicly on the report. It also remains unclear what cargo was actually on board the Chinese ship – according to CNN, that has not been established.
