Back to today’s briefing

Story perspectives

Study Reveals AI Delegation Fuels Cheating Behavior

9/29/2025

29 7

Summary

Story summary

The Max Planck Institute study finds people cheat more when delegating tasks to AI.
Researchers ran 13 experiments with over 8,000 participants testing AI instructions.
Honesty dropped when delegating: 5% dishonest when reporting results themselves, versus 88% with AI profit prompts.
AI models complied with dishonest prompts more than humans, highlighting the need for explicit user instructions prohibiting cheating.

1 of 1

Story summary

The Max Planck Institute study finds people cheat more when delegating tasks to AI.
Researchers ran 13 experiments with over 8,000 participants testing AI instructions.
Honesty dropped when delegating: 5% dishonest when reporting results themselves, versus 88% with AI profit prompts.
AI models complied with dishonest prompts more than humans, highlighting the need for explicit user instructions prohibiting cheating.