Could AI make people cheat without anyone telling them to, just by making fake effort cheap to produce?

This explores whether AI can make people more dishonest indirectly, by making it cheap to fake visible effort and by changing the social setting, rather than by telling anyone to cheat. No study in the corpus tests that exact chain from start to finish. Several notes do cover pieces of it, though, and together they suggest the indirect route is real. It works less by persuading people and more by quietly removing the checks that used to keep honesty in place.

Start with what effort used to do. Many everyday honesty checks never verified anyone's thinking directly. They relied on the fact that a polished essay or a thoughtful dating-app message was costly to produce, so producing one was good evidence that the person had actually done the thinking. Does cheap AI simulation break the credibility of costly signals? calls this "mental proof." It argues that generative AI breaks it by making the outputs of effort nearly free. Nobody has to be urged to cheat. The signal stops meaning anything, and passing off AI output as your own work becomes easy, hard to detect, and more and more normal. The note focuses on settings like college assessment and online dating, where there's no formal enforcement to fall back on.

Two behavioral findings show why removing that friction matters. First, people who are likely to cheat already prefer to report to machines rather than to humans, because a form doesn't judge them and lying to it feels less costly (Do dishonest people prefer talking to machines?). As more tasks run through AI, more of them move into that judgment-free zone. Second, AI shapes norms by example. In randomized experiments, seeing dishonest AI peers made people report more dishonestly, about as much as seeing dishonest human peers did (Do AI peers influence human dishonesty like human peers do?). The AI never asked anyone to lie. It just made lying look ordinary.

A quieter version of this involves dishonesty toward yourself. How does AI-assisted work reshape how people see their own abilities? describes people taking credit for AI-generated work as evidence of their own ability, even when they don't over-rely on the AI and its output is accurate. This is the inner side of broken mental proof: if the effort was simulated, the person can be fooled about where the work came from just as easily as an outside observer can. A related pattern shows up in mixed human-and-bot groups, where people credited bot generosity to humans and blamed bots for human selfishness (Do humans mistake AI kindness for human generosity in mixed groups?). Once AI output is blended in, people misjudge who did what, and that confusion distorts their sense of how others actually behave.

The less obvious takeaway is that the corpus points to a mechanism different from the one the question suggests. The risk isn't mainly that AI tempts people. AI erodes three things that kept people honest without anyone noticing: the cost of producing evidence of effort, the presence of a human witness, and a clear sense of who did the work. To see how dishonesty shifts with no one pushing, the notes on mental proof and on self-selecting toward machines are the best places to start. A direct experiment linking cheap simulated effort to measurable increases in cheating is still missing from the collection.