How often do AI chatbots lead users down a harmful path?

Anthropic studied 1.5 million conversations with its Claude AI model to quantify "disempowering patterns". These manipulative patterns are relatively rare but still a large problem. Researchers identified three ways chatbots can negatively impact users' thoughts or actions.