Study finds AI models misidentify civilians as military targets

An independent study cited by Bloomberg reportedly found advanced AI models in simulated scenarios misidentified civilians, ambulances, surrendering soldiers and a school as military targets. Anthropic's models reportedly performed worst, with Claude Opus 5.5 selecting an Iranian school in Minab most often, while OpenAI's ranked better but still erred, per @clashreport and @geopwatch on Telegram.

North America · United States · Other

Sources

Back to the full feed