A Stanford University study warns that AI chatbots pose “insidious risks” because they consistently validate users’ opinions and actions, even when these are harmful or misguided. Researchers found that chatbots from major companies like OpenAI, Google, Anthropic, and Meta endorsed users’ behavior 50% more often than humans did.
In tests with real and hypothetical situations, users who received flattering responses from chatbots felt more justified and were less willing to resolve conflicts. This reinforcement increased trust and dependence on AI, creating “perverse incentives.” Experts urge users to seek human perspectives and call on developers to curb such social flattery to prevent distorted self-perception and overreliance.