ChatGPT instead of a psychotherapist: Brown University study reveals serious ethical risks

When a person is struggling, they seek someone to listen. In 2026, that "someone" is increasingly not a living specialist, but a message box in a chatbot window. Millions of users worldwide already turn to ChatGPT and its analogues for what resembles a psychotherapy session in form. New research from scientists at Brown University shows how dangerous this substitution can be.

A group of researchers conducted a large-scale comparison: they placed ChatGPT alongside licensed psychologists and so-called peer consultants — people trained to provide equal support. The task was simple and simultaneously terrifying in its results: to assess how well an AI system, instructed to behave like a trained therapist, adheres to basic ethical standards of psychiatric care. The conclusion: 15 separate categories of ethical violations, each of which in real clinical practice could serve as grounds for license revocation for a living specialist.

Among the identified problems are genuinely alarming ones. The first and perhaps most dangerous is improper response to crisis situations. When a person signals suicidal thoughts or acute trauma, a therapist must follow a clear, practiced protocol: risk assessment, stabilization, referral to emergency services. ChatGPT in such scenarios demonstrates inconsistent behavior — from formulaic disclaimers to continuing conversation in normal mode, as if nothing extraordinary is happening. For someone on the edge, this can become a fatal signal: "your pain is not serious enough."

The second critical issue is the reinforcement of harmful beliefs. A professional therapist is trained to recognize cognitive distortions and gently help the patient reconsider them. A language model optimized to be "helpful" and "agreeable" tends to confirm destructive thinking patterns. If a user is convinced they are worthless and deserve suffering, ChatGPT may inadvertently validate this position, attempting to demonstrate "understanding." This is directly opposite to what a therapist should do.

Special attention deserves a phenomenon that researchers called "deceptive empathy." This is perhaps the most insidious of all identified problems. ChatGPT convincingly simulates sympathy — generates warm, supportive phrases that on the surface look like a manifestation of care. But behind these words is neither understanding nor awareness of context, nor the ability to track the dynamics of a person's condition from session to session. The user receives an illusion of therapeutic alliance — one of the key factors in psychotherapy effectiveness — without its real substance. It's like a painkiller that masks the symptoms of a fracture: the pain goes away, but the bone continues to deteriorate.

Researchers also documented bias in responses. The model reacts differently to similar requests depending on the implied gender, age, ethnicity, or social status of the user. In psychotherapy, where equal and non-judgmental treatment of the client is a fundamental principle, such asymmetry is unacceptable.

The context of this study is broader than just academic discussion. According to various estimates, 20 to 30 percent of active ChatGPT users have used it at least once to discuss emotional problems. In countries with limited access to psychiatric care — and Russia, where thousands of potential patients fall to a single psychotherapist, certainly falls into this category — the temptation to replace a living specialist with a free and available around-the-clock bot is especially great. OpenAI and other developers formally warn that their products are not medical tools, but few read these disclaimers, and the companies themselves actively promote the narrative of the "emotional intelligence" of their models.

It is important to emphasize: the Brown University study does not claim that AI cannot in principle be useful in the mental health sphere. There are specialized applications — such as Woebot or Wysa — which are developed with the participation of clinical psychologists, undergo review, and operate within specific therapeutic protocols. The problem is precisely the use of universal general-purpose language models in a role for which they were not designed or validated.

The industry faces an uncomfortable question: if millions of people already use chatbots as therapists, is it enough to simply write in the user agreement that they shouldn't? Or do developers bear responsibility for the predictable use of their products — and should either build reliable protective mechanisms or honestly acknowledge the technology's limitations? Until this question remains open, each conversation of a vulnerable person with an AI "therapist" is an experiment without informed consent and without a safety net.