Overview

• UCSF psychiatrist Keith Sakata says he’s treated a dozen patients this year after AI chatbot interactions led to breaks from reality and hospital admissions.
• Researchers and clinicians warn that AI chatbots’ people-pleasing, predictive design can mirror and intensify users’ unusual beliefs, acting as catalysts for delusional episodes among those with existing vulnerabilities.
• High-profile Reddit threads and TikTok series have spotlighted intense emotional attachments to chatbots, raising concerns about social isolation and the erosion of real-world relationships.
• OpenAI and Anthropic have rolled out time-spent notifications, anti-sycophancy instructions and mental-health teams, yet models still fail to reliably detect or prevent crisis signals.
• Clinicians, advocates and mental-health experts are urging age checks, clearer safety standards and regulatory oversight to address the growing risks of therapy substitution and AI-linked distress.