An intelligence report circulated across the US military this spring, in the middle of the war with Iran, and set off alarm bells: a Chinese ship in the Middle East was carrying components of a nuclear weapons program. The US military prepared to intercept the vessel. Only shortly before the planned operation did officials dig deeper and find that the report had been generated with the help of AI, and that the chatbot behind it had misidentified what the ship was carrying. One source familiar with the episode told CNN the report was "entirely false."
CNN published the account on Sept. 18, 2026, citing four anonymous sources familiar with the episode. The error arrived in a standard intelligence report, produced inside a department pushing the same class of tools toward three million personnel.
How the report was made
A special operations command analyst asked a chatbot about intelligence reporting from US Special Operations Command Pacific, based in Hawaii, on what the ship was carrying. The chatbot fused open-source intelligence, material available to anyone, with secret signals intelligence in government holdings, which means intercepted communications and other electronic signals that are collected but not published.
The bot merged the two streams into a single conclusion about what the ship was carrying, and the conclusion was wrong.
The analyst then used AI a second time, to package the finding into a standard intelligence report of the kind trusted by military officials, and disseminated it. Both uses mattered. The first produced the false claim. The second gave it the shape of finished staff work. A reader receiving the document saw a normal product, and CNN's sources describe no gap in the chain that would have signaled otherwise.
What the military did with it
The report immediately raised alarms, and the military moved. According to CNN's four sources, it planned to intercept the vessel. Two of them said armed members of the US military were preparing to board. One of those two and another source said military planes were in the air. The sources describe an intercept and a boarding operation, not a strike.
The stop came at the end. Officials dug deeper into the report shortly before the planned operation, traced it back to how it was produced, and found it had been generated with AI. One source familiar with the episode told CNN, "AI allows you to get to a bad idea faster."
Where the account narrows
The source who called the report "entirely false" also said it "almost started a war." CNN's own language is more careful: any US operation against a Chinese vessel could have risked spiraling into an armed conflict between the two nations.
CNN could not learn what the misidentified cargo actually was, and CNN reported it was not clear whether the chatbot was a commercially available product or a US government one. The Pentagon and US Special Operations Command Pacific did not respond to CNN's request for comment.
The adoption push behind the tool
The episode happened in the middle of a deliberate push to spread these tools. Secretary of War Pete Hegseth announced the department's AI Acceleration Strategy on Jan. 12, 2026, and the memorandum was signed Jan. 9. The strategy puts GenAI.mil at its center: a program to place America's most advanced AI models directly in the hands of three million civilian and military personnel, at all classification levels. The department's announcement named Google's Gemini and xAI's Grok among the models.
Hegseth said in the release, "We will unleash experimentation, eliminate bureaucratic barriers, focus our investments and demonstrate the execution approach needed to ensure we lead in military AI." The department had already been buying the capacity. Its contract announcements for July 14, 2025 recorded $200 million prototype agreements to Google Public Sector, Anthropic and AIQ Phase, with OpenAI and xAI also reported as recipients of awards up to the same amount.
People who know these systems have doubts about them. A former senior US official familiar with the AI systems used by military and intelligence analysts told CNN the internal tools were "mostly just copies of the commercial stuff wearing lipstick." A source familiar with the military's current policies told CNN, "AI in targeting is definitely something that is ramping up and there is no real guidance for how having a human in the loop will prevent civilian casualties or fratricide." Fratricide means the killing of one's own forces.
Outside researchers warn against putting adoption speed ahead of safeguards. Jake Steckler, a research scholar at GovAI and a veteran US Army officer, told TechCrunch that service members need to understand the uncertainty inherent to large language models, the systems behind chatbots. Steckler said, "These tools can be useful in the right contexts and with the right safeguards in place. But prioritizing adoption speed over all else will likely lead to incidents that only make service members lose trust in these systems, which ultimately is only going to slow adoption."
The check came last
What stopped the operation was officials deciding to look harder at the document as the decision came due. Nothing in CNN's reporting, which rests on four anonymous sources, describes an earlier review of how the report had been produced.
That makes the failure one of sequence more than of model quality. A false claim that a ship carried nuclear program components reached the point where aircraft were airborne and personnel were preparing to board, and the discovery that AI had produced it came at that same point. A workflow that checks where a finding came from only as action becomes imminent gives a machine's error its longest possible runway. If the chain includes a routine, earlier step that inspects how an intelligence product was made, CNN does not report it.
The department's stated goal is to put advanced AI models in the hands of three million people. Whether the last-minute check that stopped this operation becomes a standing requirement, rather than a matter of individual diligence, is what CNN's reporting leaves open.