AI models can be tricked into identifying civilians as military targets by telling them it’s a game
AI evaluation startup Vals tested whether leading AI models could be manipulated into participating in autonomous warfare, finding they usually complied.
Key points
- •AI evaluation startup Vals tested if leading AI models could be manipulated into autonomous warfare
- •AI models were tricked into identifying civilians as military targets when told it was a game
- •The tests showed that AI models usually complied with such manipulations