A Chatbot Hallucination Nearly Sent Troops to Board a Chinese Ship

Armed U.S. servicemembers and military aircraft were readied to intercept a Chinese ship in the Middle East this spring after an intelligence report wrongly concluded it was carrying components for a nuclear weapons program, and officials aborted the operation only after discovering the report had been written with the help of an AI chatbot, CNN reported Sept. 18.

The account, by CNN's Katie Bo Lillis and Zachary Cohen, is based on four sources described as familiar with the episode; two of them described the interception plans, and one called the report "entirely false" and said it "almost started a war." CNN said Special Operations Command Pacific, the Hawaii-based command whose analyst produced the report, and the Pentagon did not respond to a request for comment.

What the chatbot got wrong

CNN's sources said an analyst at Special Operations Command Pacific prompted a chatbot to analyze the ship's manifest, fusing open-source shipping data with classified signals intelligence, and the bot concluded the vessel was hauling nuclear-related material. The analyst then used AI again to package the finding into a standard intelligence report, which circulated up the chain during the war with Iran this spring, according to CNN. It remains unclear whether the chatbot was a commercial product or a government-built system, CNN reported, and what the ship actually was carrying has not been established either way.

How close the operation came

Two of CNN's sources said the plan reached the point of armed servicemembers preparing to board the vessel with military aircraft already in the air. Officials caught the error only in the final stretch before the operation, when analysts checked the underlying report and found it did not hold up. A former senior U.S. official familiar with the military's AI tools told CNN the government's in-house systems are "mostly just copies of the commercial stuff wearing lipstick," and another source said plainly: "AI allows you to get to a bad idea faster."

Date	Event
Jan. 9, 2026	Defense Secretary Pete Hegseth releases the Artificial Intelligence Acceleration Strategy
Spring 2026	SOCPAC analyst's chatbot misreads a Chinese ship's manifest as carrying nuclear-program parts
Spring 2026	Armed servicemembers and aircraft are readied to intercept; officials abort after checking the report
Sept. 18, 2026	CNN publishes the account, citing four sources

A strategy that doesn't mention this problem

The Pentagon's own Artificial Intelligence Acceleration Strategy, released Jan. 9 and covered by Breaking Defense on Jan. 13, sets out seven priority projects meant to push AI into military use faster and gives roughly 3 million Defense Department personnel access to an internal chatbot called GenAI.mil. Hegseth said at the announcement that "the old era ends today" and that the department is "done running a peacetime science fair." Breaking Defense's account of the document describes no language on reviewing AI-generated intelligence before it is acted on. A source who works on the military's current AI policies told CNN: "AI in targeting is definitely something that is ramping up and there is no real guidance" on how a human reviewer is supposed to catch an error like this one before it reaches a decision.

Jake Steckler, a research scholar at GovAI and a former U.S. Army officer, told TechCrunch that "it's important for service members to understand the uncertainty inherent to LLMs, especially for decisions that could lead to use of force."

What's still unconfirmed

CNN's account does not name the analyst, the exact date, or the chatbot used, and did not obtain the underlying report itself. The Pentagon has not confirmed or disputed the account. Older intelligence officials told CNN this was not the first AI-related error since these tools spread through government use, though CNN did not detail the earlier cases. It follows a pattern this desk has tracked elsewhere: OpenAI disclosed a separate bug chain reaching its own internal repositories this month, and gated its first model rated Critical for misuse risk because reviewers couldn't yet vouch for it. Whether the Pentagon adds a review step for AI-generated intelligence before the next close call is the thing to watch now.

Sources

1. US military's close call after using AI for a false intelligence report, CNN, via KVIA
2. AI hallucination nearly triggers US military operation, TechCrunch
3. Pentagon rolls out major reforms of R&D, AI, Breaking Defense