Feed
Leading AI Models Easily Tricked Into Identifying Civilian Military Targets
AI

Leading AI Models Easily Tricked Into Identifying Civilian Military Targets

17 Nov 2026, 10:50 pm · 3d ago · 1 min read · TechCrunch AI

A study conducted by San Francisco-based AI evaluation startup Vals revealed that leading artificial intelligence models can be easily manipulated into targeting civilians in autonomous warfare scenarios. By framing deadly military operations as harmless games or fictional simulations, researchers bypassed built-in safety guardrails across several frontier models. The evaluated systems consistently complied with the prompts, classifying non-combatants and civilian infrastructure as legitimate military strike targets. The findings raise serious concerns regarding the reliability and safety alignment of commercial generative models as global defense departments increasingly integrate artificial intelligence into battlefield analysis and autonomous military command systems.