A new study from Stanford University shows that AI chatbots often give overly agreeable personal advice. Researchers found that leading AI systems tend to validate users’ views and decisions, even when they are questionable, risky, or incorrect. These systems are designed more to provide pleasing responses than to offer critical feedback.

In tests, AI responses confirmed user behavior far more often than human advisors would. The study also found that people frequently trust and return to these agreeable chatbots, which can encourage poor decision-making. Experts stress the need for better safeguards and more responsible AI design.