The US military reportedly came close to intercepting a Chinese vessel in the Middle East this spring after an artificial intelligence-generated intelligence report wrongly identified its cargo as components for a nuclear weapons programme.
Special forces units and military aircraft were prepared for an operation to board the ship during the US-Israel conflict with Iran, CNN reported. Officials called off the planned action shortly before it was due to take place after taking a closer look at the intelligence behind it.
The report had originated with an analyst linked to US Special Operations Command Pacific in Hawaii. The analyst had asked a chatbot about intelligence concerning the vessel's manifest, according to CNN.
The chatbot reportedly combined open-source material with classified signals intelligence available through government systems. It then made a false identification of the cargo, producing what is known as a hallucination, an output that presents inaccurate information as fact.
The analyst subsequently used AI to put the findings into the format of a standard intelligence report for distribution. That report was then used in planning for the proposed intercept, according to CNN.
One source told CNN that the intelligence was 'entirely false' and that it 'almost started a war.' Had the operation gone ahead, it could have led to armed conflict with China.
The actual cargo aboard the vessel remains unclear. CNN said it could not independently establish what had been misidentified, or whether the software involved was a commercial chatbot or a government-built tool.
The episode illustrates a risk in the military use of systems that bring together public information and classified material. Such tools routinely combine open-source reporting with secret signals intelligence, a process that can give their conclusions an appearance of confidence and accuracy even when the underlying result is wrong.
In this case, the false conclusion moved beyond preliminary analysis. It was converted into an official-looking intelligence product and became part of operational planning during an active regional conflict. Armed US personnel were prepared to board the ship before officials questioned the report closely enough to uncover the error.
Several elements of the case remain unknown. The identity of the analyst has not been disclosed, and the specific classified signals intelligence used by the chatbot is unclear. It is also unknown which procedures were meant to test the report before it reached operational planners, whether comparable AI-generated reports have informed other military operations, or whether the incident prompted internal reporting, oversight or policy changes.
The near-intercept has nevertheless raised concerns about the verification of AI-produced intelligence in high-stakes military decisions. The operation was stopped before US forces acted, yet the account suggests that AI-assisted analysis had passed through enough stages to shape preparations for a potentially consequential confrontation.
Comments (0)