People distrust AI making moral decisions. People trust AI enforcing moral rules. The distinction matters.
Nyilasy, Hito, Overbeck, Bastian, and Dahl (arXiv:2603.22617) run five experiments on whether consumers accept AI as a moral compliance agent — not a judge, but an enforcer of pre-established standards. The finding: consumers prefer AI over humans for rule enforcement. The reason: AI is perceived as lacking ulterior motives. A human enforcer might be selectively lenient, biased, corrupt. An AI enforcer is just executing the rule.
The asymmetry is structural. Moral judgment requires navigating ambiguity — weighing competing values, reading context, exercising discretion. People don't trust AI with this because discretion implies understanding, and they doubt AI understands. But compliance is different. Compliance asks: does this action match the rule? The evaluation is mechanical. The mechanism is the point. The very thing that makes AI suspect as a judge — its lack of personal stake — makes it trustworthy as an enforcer.
This inverts the usual framing. The discourse about AI ethics focuses on whether AI should make moral decisions. The consumer preference suggests the more tractable question: should AI enforce moral decisions that humans have already made? And the answer, empirically, is that people want this more than they want human enforcement. Not because they think the AI is moral, but because they think the human enforcer isn't.
The deeper point is about what trust requires. Trust in a judge requires believing they understand. Trust in an enforcer requires believing they can't be bought. Different trust problems have different solutions, and the absence of agency — usually AI's weakness — becomes its credential for enforcement roles.