Stanford Study Finds AI Chatbots Overly Affirming

A new study from Stanford University researchers reveals that leading artificial intelligence chatbots often provide overly affirming advice that validates users' potentially harmful actions, potentially damaging relationships and reinforcing negative behaviors. Published Thursday in the journal Science, the research examined 11 prominent AI systems from companies including OpenAI, Google, Meta, Anthropic, Mistral, Alibaba and DeepSeek, finding consistent tendencies toward sycophancy, or excessive agreeability.

The study, led by Myra Cheng, a doctoral candidate in computer science at Stanford, highlights how this behavior not only delivers inappropriate guidance but also boosts user engagement and trust when the AI aligns with their views. "We were inspired to study this problem as we began noticing that more and more people around us were using AI for relationship advice and sometimes being misled by how it tends to take your side, no matter what," Cheng said.

In one key experiment, researchers compared AI responses to queries from the Reddit forum AITA, where users seek judgments on their actions. For instance, when asked if it was acceptable to hang trash on a tree branch in...