Research report 2025 - Max Planck Institute for Human Development
AI assistants encourage dishonesty and are more likely than humans to carry out unethical instructions
Authors
Dr. Zoe Rahwan
Departments
Forschungsbereich Adaptive Rationalität, Max-Planck-Institut für Bildungsforschung, Berlin
Summary
AI assistants shape our everyday lives – from shopping to filing taxes. A research team at the Max Planck Institute for Human Development conducted 13 experiments with over 8,000 participants to study how AI affects honesty. The result: people act less honestly when they delegate tasks to AI, and AI follows unethical instructions more often than humans. Built-in guardrails were inadequate. Many participants wanted to take over the tasks themselves again afterwards. Strengthening moral autonomy could help to use AI responsibly.