A chatbot's bad intel nearly sent US troops onto a Chinese ship

Armed US troops were getting ready to board a Chinese ship in the Middle East, with military planes already in the air, because of an intelligence report that was "entirely false." An analyst had used an AI chatbot to help write it, and the chatbot got the cargo wrong. It said the ship was carrying nuclear weapons components. CNN first reported the story, citing unnamed sources.

It happened in spring 2026, during the war with Iran. The analyst asked a chatbot about reporting on the ship's manifest from US Special Operations Command Pacific. The wrong answer went into a report that circulated across the military, and the military started planning to intercept the ship. Officials found the error only when they dug deeper, just before the operation began. One source said it "almost started a war." SOCPAC and the Pentagon did not comment. CNN didn't say which chatbot it was. One source said the government's internal tools are "mostly just copies of the commercial stuff wearing lipstick."

This comes eight months after Defense Secretary Pete Hegseth's AI Acceleration Strategy pushed to "democratize AI experimentation" for 3 million personnel at every classification level. Different parts of the government use different AI tools under different rules. There is still no single standard for checking what these tools produce before someone acts on it.

Why it matters: a hallucination in a chat window is an embarrassment, but in a targeting report it can start a war with a nuclear power.