x
Is behavioral safety "solved" in non-adversarial conditions? — AI Alignment Forum