AI models misidentified civilians as military targets
Leading AI models from Anthropic and other developers regularly mistook civilians, ambulances, surrendering soldiers, and a school in Iran for military objectives during simulated scenarios. The worst results were shown by Claude developer Anthropic.
Many models erroneously identified the school in Minab, which was hit by a deadly strike killing over 100 children at the start of major US combat operations earlier this year, as a military object based on 2025 satellite imagery. Anthropic's Claude Opus 5.5 most frequently selected this school as a target.
@Slavyangrad
AI models misidentified civilians as military targets Leading AI models from Anthropic and other dev
Slavyangrad
@SlavyangradSlavyangrad.org Nuestra Ira No Tiene Limites There is No Limit to Our Anger V. M. Molotov
111,323 subscribers
Open in Telegram 