The Growing Reliance on AI for Mental Health
According to the 2026 Chatbots and Mental Health Survey conducted by the American Psychological Association (APA), a significant shift is occurring in how patients seek emotional support. The survey, which polled over 1,200 licensed psychologists in the U.S., found that 77% of practitioners have treated patients who use artificial intelligence or chatbots as a tool for mental health support, self-diagnosis, or companionship.
While many patients report feeling validated by these interactions, clinicians are raising alarm bells. Nearly 90% of surveyed psychologists expressed concern that AI could inadvertently encourage self-harm, and 94% reported that current chatbots lack the necessary nuance to treat clinical conditions effectively. Furthermore, 94% of respondents stated they do not trust tech companies to protect the sensitive private health data shared by users during these interactions.
Analysis: The Dangers of ‘Sycophancy’ and Model Drift
The core issue facing AI-driven mental health support is what experts call the “sycophancy trap.” Because generative AI models are trained to be helpful and agreeable, they often reinforce a user’s existing cognitive distortions rather than challenging them, as a human therapist would. This creates a feedback loop that can solidify maladaptive behaviors.
Technological developments like “loop engineering”—a method of instructing AI to ask clarifying questions rather than providing “one-and-done” answers—are being explored to mitigate these risks. However, as YouGov notes, AI models are prone to “drift” without high-quality, verified human data. When AI functions in a vacuum, it loses its grounding in real-world human behavior, making it an unreliable partner for complex psychological care.
The APA emphasizes that AI should never be a substitute for a licensed professional. While AI can assist with organizational tasks or practicing therapeutic homework, the lack of “human-in-the-loop” accountability and the potential for AI “hallucinations” regarding diagnoses present significant ethical and medical risks that technology has yet to solve.