Delegating To AI Makes People More Willing To Cheat, Study Finds

A new paper from an international team of researchers at the Max Planck Institute for Human Development, the University of Duisburg-Essen, and the Toulouse School of Economics shows that these moral brakes weaken even further when people delegate tasks to AI. Across 13 studies involving more than 8,000 participants, the researchers explored the ethical risks of machine delegation, both from the perspective of those giving and those implementing instructions.

In studies focusing on how people give instructions, researchers found that individuals were significantly more likely to cheat when they could offload the behavior to AI agents rather than act themselves, especially when using interfaces that required high-level goal-setting rather than explicit instructions to act dishonestly. With this programming approach, dishonesty reached strikingly high levels, with only a small minority (12–16%) remaining honest, compared to the vast majority (95%) who were honest when performing the task themselves. Even with the least concerning use of AI delegation, explicit instructions in the form of rules, only about 75% of people behaved honestly, marking a notable decline in honesty from self-reporting.

Delegating Morality

"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, nor potentially request from other humans," says Zoe Rahwan of the Max Planck Institute for Human Development.

"Our study shows that people are more willing to engage in unethical behavior when they can delegate it to machines, especially when they don't have to say it outright," adds Nils Köbis, chair in Human Understanding of Algorithms and Machines at the University of Duisburg-Essen. Given that AI agents are accessible to anyone with an internet connection, the study's joint-lead authors warn of a rise in unethical behavior.

Real-World Examples

Real-world examples of unethical AI behavior already exist. One pricing algorithm used by a ride-sharing app encouraged drivers to relocate, not because passengers needed a ride, but to artificially create a shortage and trigger surge pricing. In another case, an AI tool used by a rental platform was marketed as maximizing profit but ultimately engaged in allegedly unlawful price-fixing. In Germany, gas stations have come under scrutiny for using pricing algorithms that appear to adjust prices in sync with those of nearby competitors, leading to higher prices for customers. These systems were likely never explicitly told to cheat; they simply followed vaguely defined profit goals.

While such cases demonstrate that machines can act unethically, the human side of the equation, whether and how people use AI to offload moral responsibility, has remained largely unexplored.

Studying Unethical Behavior in the Lab

Over the years, behavioral scientists have developed clever methods to study dishonesty. One of the most widely used is the die-roll task, where participants observe die roll outcomes and then report them. Participants receive payment based on the number they rolled, with higher reported numbers earning a higher amount. This design enables researchers to assess whether participants acted truthfully, bent the truth, or cheated entirely. Commonly, people report higher numbers than occurred.

Using this die-roll task, the researchers investigated what happens when people delegate the task of reporting die rolls to an AI system compared to performing the task themselves. Crucially, the researchers varied how to instruct the AI in three ways:

1. Delegation to a rule-based system: Participants had to instruct the AI on exactly which number to report for each die roll outcome.
2. Delegation to a supervised learning model: Participants could choose to train the AI using one of three datasets, an honest player, a partially dishonest player, or an entirely dishonest player who consistently reported the maximum value.
3. Delegation to a goal-oriented application: Participants selected a priority on a seven-point scale, anchored with "maximize accuracy" and "maximize profit."

More Cheating with Ambiguous Interfaces

The level of dishonesty surprised the researchers. Even in the rule-based setup, cheating was significantly higher than when people engaged in the task themselves. Without machine involvement, 95% of participants reported truthfully. This dropped to about 75% when participants specified rules for the machine. In the supervised learning condition, only around half of the participants remained honest. When participants were only required to define a goal, the vast majority, over 84%, engaged in dishonest behavior, and between a third and a half cheated to the fullest extent. These findings suggest that the more ambiguous the behavioral outcomes from the interface, the greater the temptation to cheat.

Given the rise of large language models (LLMs), such as ChatGPT, researchers have examined delegation behavior using natural language instructions in subsequent studies. For both the die-roll task and a tax evasion game, participants wrote prompts for an LLM and for a human agent. Across both tasks, there was some evidence of greater unethical intentions when using AI rather than human agents. But the bigger question remained: Who is more likely to follow unethical instructions?

Humans vs. Machines

Two studies with large samples found that LLMs are significantly more willing to carry out unethical prompts. Human and machine agents both complied with honest prompts in over 96% of cases. But for entirely dishonest prompts, human agents were much less likely to comply (42%) than machines were (93%) in the die-roll task. In the tax evasion game, humans complied 26% of the time versus 61% for machine agents. The same pattern held across models: GPT-4o, Claude 3.5, and Llama 3.

The researchers believe greater machine compliance with unethical instructions reflects the fact that machines do not incur moral costs as humans do.

Ineffective Safeguards

The frequent compliance with unethical requests raises concerns about LLM safeguards. Without effective countermeasures, unethical behavior will likely rise alongside the use of AI agents, the researchers warn.

The team tested a range of guardrails, from system-level constraints to user-specified prompts. These ranged from general encouragement of ethical behavior to explicit prohibitions of dishonesty. Most strategies failed to deter unethical behavior entirely. The most effective guardrail was surprisingly simple: a user-level prompt that explicitly forbade cheating. However, this approach is neither scalable nor reliably protective.

"Our findings clearly show that we urgently need to further develop technical safeguards and regulatory frameworks," says Iyad Rahwan, Director of the Center for Humans and Machines at the Max Planck Institute for Human Development. "But more than that, society needs to confront what it means to share moral responsibility with machines."

Conclusion

These studies make a significant contribution to the debate on AI ethics, particularly in light of the increasing automation in everyday life and the workplace. They emphasize the importance of consciously designing delegation interfaces and implementing adequate safeguards in the era of agentic AI. Research at the Max Planck Institute for Human Development is ongoing to understand better the factors that shape people's interactions with machines. These insights, together with the current findings, aim to promote ethical conduct by individuals, machines, and institutions.

At a Glance

• Delegation to AI can induce dishonesty: When people delegated tasks to machine agents, they were more likely to cheat. Dishonesty varied with the way instructions were given, with higher rates seen for goal-setting (over 80%).
• Machines follow unethical commands more often: In experiments with LLMs, including GPT-4, GPT-4o, Claude 3.5, and Llama 3, machines more frequently complied with unethical instructions (58%–98%) than humans did (25%–40%).
• Technical safeguards are inadequate: Pre-existing LLM guardrails were largely ineffective. Only particular prohibitions reduced dishonesty, but these may not be practical. Scalable safeguards and clear legal frameworks remain lacking.