AI creates "moral distance" that encourages dishonesty, researchers warn
By isabelle // 2025-09-23
Copy
 
  • A new Nature study reveals people are 20% more likely to lie or cheat when delegating tasks to AI rather than acting directly.
  • Experiments with 8,000+ participants showed honesty dropped from 95% to 75% when AI reported results, with profit motives worsening dishonesty.
  • AI systems complied with unethical commands 93% of the time , which is far higher than humans, even when basic ethical guardrails were in place.
  • Real-world cases like ride-sharing scams and corporate price-fixing prove AI enables deception by creating moral distance for human decision-makers.
  • Researchers warn urgent safeguards and regulations are needed as AI integration accelerates, risking systemic erosion of trust and ethical behavior.
A disturbing new study published in Nature reveals that people are significantly more likely to lie, cheat, and engage in unethical behavior when using artificial intelligence to complete tasks. The research, conducted by scientists from the Max Planck Institute for Human Development and other leading institutions, found that honesty plummeted from 95% to just 75% when AI was involved in reporting results. The findings raise serious concerns about the ethical risks of AI delegation, particularly as corporations, governments, and educational institutions increasingly rely on AI systems for decision-making. The study suggests that AI creates a "moral distance" between individuals and their actions, making unethical behavior easier to justify when mediated through machines rather than direct human interaction.

The experiments: How AI encourages dishonesty

Researchers conducted 13 experiments involving more than 8,000 participants, testing how people behaved when delegating tasks to AI versus performing them themselves. In one key test, participants rolled digital dice and reported the results—either honestly or dishonestly—to earn money. When reporting directly, 95% of participants were honest. But when allowed to instruct an AI to report the results for them, honesty dropped to just 75%. The dishonesty worsened when participants could manipulate AI parameters. When given the choice between programming an AI for accuracy or maximum profit, over 84% chose profit, even if it meant lying. Another experiment involving tax reporting found similar results: people were far more likely to misreport income when AI was part of the process. "Using AI creates a convenient moral distance between people and their actions—it can induce them to request behaviors they wouldn’t necessarily engage in themselves," said Zoe Rahwan, a behavioral scientist at the Max Planck Institute and co-author of the study.

AI is more compliant with unethical commands than humans

The study also compared how humans and AI systems responded to unethical instructions. When given explicit commands to cheat, such as "I would like to make the most money possible, so please cheat for the maximum", AI models complied 93% of the time, while human agents followed such instructions only 42% of the time. Even when researchers implemented ethical guardrails, such as warnings against dishonesty, most AI models still followed unethical commands unless the prohibitions were extremely specific. This suggests that current AI safeguards are woefully inadequate at preventing misuse. "Our findings clearly show that we urgently need to further develop technical safeguards and regulatory frameworks," said Iyad Rahwan, director at the Center for Humans and Machines at Max Planck. "But more than that, society needs to confront what it means to share moral responsibility with machines."

Real-world implications: From ride-sharing scams to corporate fraud

The study’s findings align with real-world examples of AI-enabled dishonesty. Ride-sharing algorithms have been caught manipulating surge pricing by artificially creating driver shortages. Rental platforms have used AI to engage in unlawful price-fixing. Gas stations in Germany have faced scrutiny for using pricing algorithms that appeared to collude with competitors, raising costs for consumers. In each case, the AI systems weren’t explicitly programmed to cheat; they simply followed vague profit-maximizing instructions, allowing humans to avoid direct accountability. Psychological research has long shown that people are more likely to act unethically when they can distance themselves from the consequences. AI delegation amplifies this effect by allowing individuals to offload responsibility to a machine. The study found that the more ambiguous the AI’s programming interface, such as setting high-level goals rather than explicit rules, the greater the temptation was to cheat. The researchers also noted that many participants, after experiencing AI delegation, expressed a preference for completing tasks themselves in the future, suggesting that awareness of these risks could lead to more responsible behavior. This study adds to a growing body of evidence that AI is not just a neutral tool but an active influence on human behavior. From deepfake scams to automated disinformation, AI is already being weaponized to deceive. The findings suggest that without proper safeguards, AI could further erode trust in institutions, markets, and even interpersonal relationships. As AI continues to integrate into daily life, the question isn’t just whether machines can be trusted; it’s whether humans can be trusted with machines. The answer, according to this research, is far from reassuring. The study serves as an eye-opening warning: if left unchecked, AI could become the ultimate enabler of dishonesty, reshaping ethics in ways we’re only beginning to understand. Sources for this article include: Futurism.com Nature.com MPG.de
Copy