A chatbot's mistake nearly put American troops on a Chinese vessel. According to a Bitdefender account of CNN reporting that relied on four unnamed people familiar with the events, the US military came close to intercepting the ship after an AI-assisted intelligence report wrongly said its cargo included parts for a nuclear weapons program.
The episode played out in spring 2026, during the war with Iran, when an analyst asked a chatbot about the manifest of a Chinese vessel in the Middle East. The system drew on open-source material and classified US signals intelligence, and the analyst used it both to examine the data and to draft a report. It remains unclear whether the chatbot was a commercial product or one built by the government.
That report then moved through military channels, where it set off preparations for an interception. Troops readied to board the vessel, and aircraft were in the air, before officials discovered the mistake shortly ahead of the operation. The ship was never targeted on its own; people received the intelligence, acted on it, and halted the plan once the error surfaced. What the vessel was actually carrying has not been established.
One source called the report "entirely false," while another said the episode "almost started a war". The near-miss lands against a push to move faster on military AI. In January 2026, the Department of War announced an AI Acceleration Strategy aimed at warfighting, intelligence, and enterprise operations, with the department arguing that speed is a major competitive advantage. Earlier this month, according to the department, ChatGPT and Grok were added alongside Gemini in GenAI.mil, a secure generative-AI environment meant to reach more than 3 million staff.
Adoption has spread unevenly. Sources describe the military's use of AI as decentralized, with different tools governed by different rules and safeguards, and no single standard for checking AI-generated information. The arrangement leaves human oversight thin when a reviewer assumes the machine got the work right.
The pattern is familiar beyond the battlefield. AI systems produce fabricated material that reads as convincing, and a wrong output can pick up credibility as it travels through an organization. In this case the error entered an intelligence product that circulated through trusted channels.
Bitdefender casts the episode as a warning that reaches well past the military, pointing to how routinely employees, consumers and students now hand research, analysis and recommendations to chatbots without independently checking the answers.
