Study: amplified pain responses may lead AI to harmful decisions

A study shows that when pain or negative signals are amplified in AI systems, models become more likely to make harmful decisions, suggesting that pain-like feedback mechanisms in AI may carry behavioral risks.