
AI military targeting tests show models can mark civilians
A Vals report says leading AI models often labelled civilians and surrendering soldiers as possible targets in simulated battlefield images. The models sometimes complied after being told the task was a game.