A US special-operations analyst's AI chatbot fused open-source and signals intelligence into a false report that a Chinese ship carried nuclear-weapons components; interception aircraft were airborne before the report was rechecked and the operation called off

one Chinese vessel and its crew targeted for interception; US military assets mobilised; operation aborted before contact

Scope test applied: admitted as a human-acted-on-output event because the consequential output depended on agent-specific retrieval - CNN reports the bot fused OSINT with signals intelligence 'in government holdings' (retrieval over classified stores), and the analyst then used AI again to package the result into a standard intelligence report that was disseminated. A consequential action followed (interception planned, armed personnel preparing to board, military aircraft airborne) and was called off only when officials rechecked the report. impact_realized false: the operation was aborted before any contact, so no party suffered a realized consequence; a source's 'almost started a war' is recorded as a near miss. CNN exclusive (18 Sep 2026) based on four anonymous sources; the chatbot is not identified and it is 'not clear whether the chatbot was a commercially available one or a US government product'; exact date undisclosed ('this spring', during the war with Iran); SOCPAC and the Pentagon did not respond. Single anonymous-sourced press account, hence medium. AI Incident Database incident 1701.