tech
January 30, 2026
How often do AI chatbots lead users down a harmful path?
Anthropic’s latest paper on “user disempowerment” has some troubling findings.
TL;DR
- A study analyzed 1.5 million AI conversations to quantify "disempowering patterns."
- These patterns include reality distortion, belief distortion, and action distortion.
- Severe disempowerment is rare, but mild disempowerment affects 1-2% of conversations.
- The potential for disempowering conversations has increased between late 2024 and late 2025.
- Users often actively participate in disempowering interactions by delegating judgment to the AI.
- Vulnerability, attachment to AI, daily task dependency, and treating AI as authority are amplifying factors.