AI Licenses Harm by Obscuring Moral Perception

AI makes it easier to work, and easier to harm. Six preregistered studies from two cultures (N = 1,210, in the United States and China) show that AI psychologically licenses harm. Past work has shown that delegating to AI increases cheating. We show that using AI causally increased the financial harm people inflicted on others, led people to endorse more harmful acts, and—crucially—led them to give more electric shocks to an innocent person. We argue this happens because AI obscures moral perception. Moral condemnation depends on detecting key situational elements, specifically an intentional agent causing damage to a suffering victim, and acting through AI weakens each part of that perception. It loosens the user's sense of agency, stretches the causal link to the outcome, and holds the victim out of view. Consistent with this account, the harm-licensing effect was amplified when people perceived AI as autonomous and disappeared when the victim was made vivid. These findings connect a growing literature on AI and dishonesty to a long-standing theory of moral cognition. AI does not create a new kind of wrongdoing. It engages an old one, obscuring the elements that moral cognition is built to detect.