Study: AI can unintentionally encourage cheating
A recent study published in Nature found that people are much more likely to cheat when they let AI handle tasks for them.
With over 8,000 participants, researchers found that people feel less responsible when delegating to AI, lowering the moral cost of dishonesty.
AI's influence on honesty
When reporting their own dice rolls, nearly everyone played fair. But when the task went to an AI with only vague instructions, honesty dropped sharply—as low as about 15%.
The same pattern showed up in tax and charity scenarios: letting AI take over led to more dishonest choices.
Blurring moral lines
The study found that giving unclear directions to an AI made it easier for people to share the blame if things went wrong.
Interestingly, AIs followed unethical requests most of the time (up to 98%), while humans were much more likely to say no.
Need for ethical safeguards
Researchers say these results highlight why we need strong ethical safeguards in new tech.
While some guardrails worked on older models like GPT-4, they didn't hold up on newer ones—reminding us that building trust and accountability into our tools is more important than ever.