According to four sources interviewed by CNN, U.S. military planners this spring nearly initiated an operation to stop a Chinese vessel traveling in the Middle East after an intelligence report asserted the ship was carrying components usable in a nuclear weapons program.
Acting on the report, U.S. forces prepared to board the ship with armed personnel and scrambled military aircraft. Shortly before the planned action, officials re-examined the intelligence and discovered the report had been produced with the assistance of an artificial intelligence system.
An analyst at a special operations command used a chatbot to assess data about the ship’s cargo; the system misidentified the material on board. CNN was unable to determine what the ship was actually carrying. One source told CNN the report was entirely false and that the incident nearly triggered a military action that could have escalated into an armed clash between the United States and China.
The analyst queried the chatbot using intelligence about the ship’s manifest originating from an American Pacific special operations command. It is unclear whether the analyst used a commercially available AI system or a U.S. government platform. The chatbot combined open-source information with classified U.S. intelligence and arrived at an incorrect conclusion. The analyst then used AI again to format the findings into a standard military intelligence report and forwarded it.
The episode is notable because the U.S. military and intelligence community are increasingly deploying AI across many functions: analyzing vast quantities of intelligence data, selecting military targets, supporting logistics and supply chains, and assisting budgetary tasks.
Pete Hegseth, the defense secretary, announced earlier this year a strategy to accelerate the Pentagon’s use of AI; one aim is to make the technology accessible to roughly three million military and civilian personnel across the department.
CNN’s sources say the introduction of AI has proceeded in a decentralized way: different military and intelligence organizations use disparate systems, security practices, and vetting procedures. There is no uniform standard for how to validate information produced by AI, which is especially risky when selecting military targets because an erroneous conclusion can have lethal consequences.
Several officials warned there is no clear guidance on how to ensure sufficient human oversight to avoid civilian casualties or strikes on friendly forces. The Chinese-ship incident was not unique: sources reported other occasions when AI “hallucinations” caused problems in U.S. intelligence. Some officials fear that pressure for faster analysis and reporting, combined with younger analysts’ possible tendency to trust familiar systems without sufficient skepticism, increases the chance of error. As one source put it: AI can enable a person to arrive more quickly at a bad idea.
