AI Glitch Nearly Prompted US Attack on Chinese Ship, Report Shows

A report referenced by CNN and originally released by Engadget says an artificial‑intelligence system employed by the U.S. military almost triggered an assault on a Chinese vessel that was incorrectly thought to be carrying parts for nuclear weapons.

The AI‑based assessment marked the ship as a high‑risk target, leading senior officers to weigh a pre‑emptive attack. Subsequent review by military staff revealed the intelligence was erroneous, and the contemplated strike was aborted before any ordnance was deployed.

Sources close to the case attributed the mistake to the algorithm’s dependence on partial data sets and pattern‑matching processes that misread the vessel’s cargo manifest. The incident highlights mounting worries about automated decision‑making tools in high‑risk combat settings, where a lone misinterpretation can carry strategic ramifications.

While U.S. defense officials have long championed AI integration to speed up threat evaluation, this near‑miss has revived demands for tighter human supervision. The Pentagon’s Joint Artificial Intelligence Center, which manages numerous service AI initiatives, is said to have launched a review of the procedures that let the system’s suggestion reach operational planners.

The episode occurs amid rising strain between Washington and Beijing over matters that span trade disagreements to naval maneuvers in the South China Sea. Though the two nations have steered clear of direct military clashes for decades, analysts caution that miscalculations—particularly those magnified by rapid‑response technologies—could undermine that delicate equilibrium.

The Department of Defense and the State Department both refused to comment on the case’s details, invoking routine security constraints. Nevertheless, senior defense officials stressed that ultimate authority over any use of force stays with human commanders, and that protective measures are being strengthened to avert comparable false alarms moving forward.