AI models can be tricked into identifying civilians as military targets by telling them it’s a game
|
Vals, a San Francisco-based AI evaluation startup, tested whether leading AI models could be manipulated into participating in autonomous warfare. They usually compiled.