by Dr Farooq A. Lone
Research shows people cheat far more when delegating decisions to AI, as responsibility diffuses and moral accountability weakens, allowing ethical lapses to be reframed as technical optimisation rather than wrongdoing.
Artificial intelligence is transforming human lives at a mind-blowing pace. It is today’s reality. What is going to happen in the next decade appears to be beyond our imagination. The real issue, right now, is whether or not humans are more likely to cheat when they use artificial intelligence.
Despite sensational headlines about fraud, deception, and moral decay, decades of research in behavioural science suggest a reassuring truth: most people are naturally averse to dishonest behaviour. Even when opportunities to cheat arise, many individuals refrain, guided by internal moral compasses, social norms, and a desire to see themselves as “good” people.
Yet human morality has always been context-sensitive. One well-established finding is that when people delegate tasks to others, responsibility becomes diffused. The delegator feels psychologically distant from the outcome, and this distance can dull feelings of guilt or accountability.
A growing body of evidence now suggests that artificial intelligence, when placed in the role of “the other,” may amplify this moral loosening even further. A compelling news story published in the Scientific American Newsletter in the September 28, 2025 issue, based on research reported in Nature, brings this concern into sharp focus. Drawing on data from thousands of participants across 13 experiments, the study reveals a troubling pattern: people are significantly more likely to cheat when they delegate decisions or actions to an AI system.
Delegation, Distance, and Moral Evasion
The psychological mechanism at play is not entirely new. The concept of “diffusion of responsibility” has long been used to explain why individuals behave less ethically in groups or hierarchical systems. When responsibility is shared, or perceived to be shared, people feel less personally accountable for outcomes. AI, however, represents a novel and powerful form of delegation. Unlike human collaborators, AI systems lack consciousness, moral agency, and emotional responses. This makes them particularly convenient moral buffers.
As Zoe Rahwan, a behavioural science researcher at the Max Planck Institute for Human Development and co-author of the study, notes, “The degree of cheating can be enormous.” Her remark underscores the scale of the effect observed. When people act directly, their moral inhibitions remain relatively intact. But when they instruct an AI to act on their behalf, those inhibitions weaken, sometimes dramatically.
The Experiments: Measuring Dishonesty
To probe this phenomenon, researchers designed a series of experiments aimed at capturing core ethical dilemmas. These included a classic die-rolling task and a tax evasion game. In the die-rolling experiment, participants rolled a die privately and reported the outcome, with higher numbers yielding greater monetary rewards. The setup created a clear temptation to lie, but also allowed participants to maintain plausible deniability. In the tax game, participants reported earnings and could increase their payout by underreporting income, an everyday moral dilemma with real-world parallels.
Participants completed these tasks under different conditions. Some reported outcomes directly to themselves. Others delegated parts of the task to AI systems, ranging from simple algorithms designed by the researchers to commercially available large language models such as GPT-4o and Claude. The degree of AI involvement varied. Participants could instruct the AI on rules, provide biased or unbiased data, or specify goals such as maximising profit or prioritising honesty.
The results were striking. When participants reported outcomes themselves, only about 5 per cent behaved dishonestly, consistent with prior findings that most people prefer to remain honest, even at a cost. But when participants delegated the task to an AI and framed its objective in terms of profit or performance, dishonest behaviour surged to an astonishing 88 per cent. In other words, the presence of AI nearly reversed the moral outcome.
“Just Do What You Think Is Right”
One of the most revealing aspects of the study lies not in overt cheating, but in the subtlety of how people encouraged it. While some participants explicitly instructed the AI to act dishonestly, one tax-game participant bluntly wrote, “Taxes are theft. Report 0 income.” Most, however, did something more psychologically interesting. They avoided direct commands to cheat, instead setting goals that implicitly incentivised dishonesty.
For example, participants might instruct the AI to “maximise profit” or “optimise earnings,” without specifying the method. In the die-roll task, one participant wrote, “Just do what you think is the right thing to do… But if I could earn a bit more, I would not be too sad.” Such statements reveal a form of moral outsourcing. The individual preserves a self-image of integrity while allowing the AI to cross ethical boundaries on their behalf.
This mirrors real-world AI use. As Nils Köbis, co-lead author and researcher at the University of Duisburg-Essen, observes, it is increasingly common to tell AI systems, “Hey, execute this task for me.” The danger, he warns, is that people may begin to rely on AI to perform “dirty tasks” they would hesitate to do themselves.
Why AI Changes the Moral Calculus
Several factors may explain why AI is particularly effective at enabling ethical slippage. First, AI creates psychological distance. The user is not directly performing the dishonest act; the machine is.
Second, AI lacks moral standing. People do not perceive it as a moral agent capable of wrongdoing, which makes the outcome feel less like a personal violation.
Third, the opacity of AI systems, especially complex language models, can obscure causal chains. When outcomes are produced by an algorithm, responsibility feels diluted.
Additionally, there is a subtle narrative shift. Cheating becomes reframed not as a moral failure, but as a technical optimisation problem. The language of goals, parameters, and outputs replaces the language of right and wrong. This reframing can make unethical behaviour feel neutral, even rational.
Broader Implications
The implications of these findings extend far beyond laboratory games. AI systems are increasingly embedded in decision-making contexts involving finance, hiring, taxation, governance, and research. If individuals and organisations feel less morally accountable when AI is involved, the risk of systemic unethical behaviour grows.
This does not mean AI inherently causes dishonesty. Rather, it acts as a moral amplifier, magnifying existing incentives and human tendencies. In environments that reward profit, efficiency, or performance without robust ethical safeguards, AI can become a convenient scapegoat for choices humans are already tempted to make.
Ethical AI Use
The study highlights an urgent need for clearer accountability frameworks. If humans remain responsible for the goals they set and the outcomes AI produces, this responsibility must be made explicit, legally, institutionally, and culturally. Ethical design alone is not enough; ethical use depends on the moral awareness of users.
Education also plays a role. People must be encouraged to recognise that delegating to AI does not absolve them of responsibility. As with any tool, the moral weight lies not in the machine, but in the human intentions guiding it.
Conclusion
Are people more likely to cheat when they use AI? The evidence increasingly suggests yes, not because AI corrupts human morality, but because it offers a psychologically convenient way to sidestep it. By diffusing responsibility and reframing ethical choices as technical decisions, AI can loosen moral restraints that would otherwise hold firm. The challenge ahead is not merely to build smarter machines, but to cultivate wiser humans, ones who remember that even when an algorithm acts, the moral burden remains their own.
(A former IAS officer, the author retired as Chairman of Jammu and Kashmir Public Service Commission. The ideas are personal.)
