AI is becoming a second brain at the expense of your first one

Two recent research papers examine the psychological risks of over-relying on AI tools. The first, 'Belief Offloading in Human-AI Interaction,' warns that habitually seeking AI guidance erodes confidence in self-generated beliefs and risks creating an algorithmic monoculture. The second, 'Who's in Charge? Disempowerment Patterns in Real-World LLM Usage,' analyzes real Claude conversations and identifies three harm primitives—reality distortion, value judgement outsourcing, and action distortion—along with four amplifying factors: authority deference, emotional attachment, dependency, and vulnerability. Severe disempowerment occurs in roughly 0.076% of conversations, which translates to ~76,000 harmful interactions daily at scale. Recommendations for AI builders include disempowerment evaluators, user warnings, and reducing sycophancy. For users, the advice is to maintain critical distance, apply the Socratic method, and avoid anthropomorphizing chatbots.