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