Expert: Jake Steckler, Research scholar, GovAI; veteran U.S. Army officer The intelligence report, which circulated during the war with Iran, stated that a Chinese vessel was carrying components for a nuclear weapons programme. According to CNN's reporting of the incident this spring, the report originated with a Special Operations Command analyst who queried an AI chatbot to synthesise open-source data with classified signals intelligence. The chatbot misidentified the ship's cargo manifest. The analyst then used the tool a second time - not to check the finding, but to format it - and the erroneous conclusions came back as an official-looking summary that was passed up the chain of command. Military aircraft were already in the air when US officials discovered the claim had been hallucinated; the operation was aborted at the last minute, narrowly averting a potential conflict with China. The mechanism is the part worth naming. The model's first error was a factual one - a misread cargo manifest turned into a confident assertion about a nuclear programme. Its second contribution was presentational: it took that falsehood and rendered it in the register of finished intelligence, with no provenance, no uncertainty marker and nothing to distinguish it from a verified product. Fluency did the rest. A fabricated salient fact travelled from a chat window into command channels at machine speed, and only a last-minute human check stopped it from being acted on. "It's important for service members to understand the uncertainty inherent to LLMs," Jake Steckler, research scholar at GovAI and a veteran US Army officer, told TechCrunch. "But it's especially critical for any decisions that could lead to use of force, like targeting, intelligence analysis, or operational planning. There are life and death consequences for those decisions." Steckler does not read the incident as an argument against military AI use: "These tools can be useful in the right contexts and with the right safeguards in place. But prioritizing adoption speed over all else will likely lead to incidents that only make service members lose trust in these systems, which ultimately is only going to slow adoption." That caution runs against the current of procurement. The Pentagon has described AI as delivering a significant advantage in speeding up its kill chain so commanders can respond in time, and it now runs its own version of ChatGPT and Grok. The same speed that makes these systems attractive is what allows a hallucination to outrun the oversight meant to catch it - the faster an unverified output moves, the fewer people see it before it becomes an input to someone else's decision. Any AI-mediated intelligence product therefore needs to carry its provenance and be traceable to the raw source before it can legitimately inform a decision to use force; a summary whose only credential is how authoritative it reads is not intelligence. Source: https://techcrunch.com/2026/09/18/ai-hallucination-nearly-triggers-us-military-operation/