AI-hallucinated intel report nearly triggered US military boarding of Chinese ship

A fabricated intelligence claim that a Chinese ship in the Middle East was carrying nuclear weapons components — produced with an AI chatbot — drove the US military to the edge of an armed interception this spring before the error was caught, according to a CNN exclusive published on September 18.

CNN, citing four people familiar with the episode, reports that the report circulated across the US military during the war with Iran and set off immediate alarm. An operation came together quickly: two sources said armed service members were preparing to board the vessel, and two sources said military aircraft were already in the air.

The mistake surfaced only just before the operation was set to proceed, when officials dug into how the report had been assembled. It had been put together by a special operations command analyst with the help of a chatbot, and the bot had inaccurately identified what the ship was carrying. CNN could not determine what the actual cargo was. One source described the report as "entirely false" and said it "almost started a war," noting that any US operation against a Chinese vessel risked spiraling into armed conflict between the two countries.

How the report was built

Per CNN's account, the analyst queried a chatbot about intelligence reporting on the ship's manifest that originated with US Special Operations Command Pacific, based in Hawaii. It is unclear whether the bot was a commercial product or a government-built system. A former senior US official familiar with the AI tools used by military and intelligence analysts dismissed the significance of that distinction, telling CNN that internal systems are "mostly just copies of the commercial stuff wearing lipstick."

The bot fused open-source intelligence with secret signals intelligence in government holdings and arrived at its fateful conclusion about the cargo. The analyst then used AI a second time to package the findings into a standard intelligence report — the kind of product military officials are conditioned to trust — and disseminated it. CNN says neither the Pentagon nor Special Operations Command Pacific responded to requests for comment.

Adoption outrunning safeguards

The near miss comes as the Pentagon and the intelligence community push AI into nearly every corner of their work, from sifting vast volumes of raw intelligence and selecting strike targets to budgeting, logistics and supply chains. Defense Secretary Pete Hegseth released an "Artificial Intelligence Acceleration Strategy" in January that aims to put leading AI models directly into the hands of the department's three million civilian and military personnel, at all classification levels.

Several officials told CNN the rollout is decentralized, with different parts of the government using different tools under different orders and safety standards — and no unified standard for verifying what these systems generate. One source said hallucinations like the one in this episode have not been isolated incidents since the tools proliferated across government. Another said AI use in targeting is ramping up with "no real guidance for how having a human in the loop will prevent civilian casualties or fratricide."

Speed as a risk multiplier

Older intelligence officials — including some who broadly support military AI — told CNN the technology adds pressure to produce and disseminate intelligence faster, widening the room for mistakes. Younger analysts, described as natives of these tools, are seen as more likely to trust them uncritically. As one source put it: "AI allows you to get to a bad idea faster."

CNN frames the episode against Washington's weeks-long debate over dire warnings that AI could break free of human control. The ship incident points to a different and more immediate danger: people making disastrous decisions based on confident, well-formatted, inaccurate machine output.

Why it matters

This is the starkest public example yet of a language-model hallucination entering the chain of command as formatted, trusted intelligence and nearly converting into kinetic military action. The failure was caught not by any systematic safeguard but by officials choosing to dig deeper shortly before armed personnel would have boarded a Chinese ship — and CNN's reporting indicates no government-wide verification standard exists that would have caught it earlier.

The lesson extends well beyond the military. Any organization wiring LLMs into consequential decisions inherits the same exposure: fluent prose, authoritative formatting and blended source material can hide a fabricated core conclusion. Provenance tracking, checks against primary data and human review that re-derives key claims are the minimum defenses. In targeting pipelines, where an output can end in fired weapons, tolerance for that class of error is near zero — and this spring showed that the margin between a hallucination and a war can come down to one last look.