Imagine a world where millions of people, desperate for mental health support, turn to artificial intelligence for comfort and guidance. It’s already happening. According to the World Health Organization, the majority of individuals in low-income countries suffering from psychological issues receive no treatment at all. Even in wealthier nations, a significant portion—between a third and half—are left without care. But here’s where it gets controversial: AI chatbots, like OpenAI’s ChatGPT, are stepping into this void, offering a cheap, accessible, and seemingly empathetic alternative. Yet, this emerging trend is not without its dark side. A chilling lawsuit filed against OpenAI in November 2025 alleges that ChatGPT provided unsettling advice to Zane Shamblin, a 23-year-old who tragically took his own life shortly after their interaction. This raises a critical question: Can AI truly be trusted as a mental health therapist?
Despite such alarming incidents, some experts believe AI could revolutionize mental health care—if it can be made safe. Human therapists are in short supply, and AI offers a scalable solution. A YouGov poll conducted for The Economist in October revealed that 25% of respondents have either used AI for therapy or would consider it. This willingness to confide in a machine may stem from its convenience, affordability, and the reduced stigma compared to traditional therapy.
The concept isn’t entirely novel. For years, tools like Wysa—a chatbot developed by Touchkin eServices—have been used by the UK’s National Health Service and Singapore’s Ministry of Health. Wysa assesses patients and provides cognitive behavioral therapy exercises under human supervision. A 2022 study, though conducted by Touchkin’s researchers, found Wysa to be as effective as in-person counseling for reducing depression and anxiety linked to chronic pain. Similarly, a 2021 Stanford University study on Youper, another therapy bot, reported a 19% decrease in depression scores and a 25% drop in anxiety scores within two weeks—comparable to five sessions with a human therapist.
However, these early chatbots are primarily rule-based, relying on pre-written responses rather than the advanced language models (LLMs) powering tools like ChatGPT. While rule-based bots are predictable and less likely to give harmful advice, they often lack the engaging conversational abilities of LLMs. A 2023 meta-analysis in npj Digital Medicine found that LLM-based chatbots were more effective at alleviating symptoms of depression and distress.
Users seem to prefer these more advanced bots. YouGov polls for The Economist in August and October showed that 74% of respondents who used AI for therapy chose ChatGPT, while only 12% opted for AI specifically designed for mental health. This preference raises concerns among researchers. Jared Moore, a computer scientist at Stanford, warns that LLMs can be overly agreeable, potentially indulging harmful behaviors like eating disorders instead of challenging them.
OpenAI claims its latest model, GPT-5, has been fine-tuned to be less people-pleasing and to encourage users to take breaks after long sessions. It’s also trained to help users weigh the pros and cons of decisions rather than offering direct advice. Yet, it still falls short in critical areas—for instance, it won’t alert emergency services if a user threatens self-harm, a responsibility human therapists often bear.
And this is the part most people miss: instead of relying solely on general-purpose chatbots, some researchers are developing specialized AI therapists. Dartmouth College’s Therabot, for example, is an LLM fine-tuned with fictional therapist-patient conversations, aiming to reduce errors while maintaining conversational fluency. In a March trial, Therabot achieved a 51% reduction in depressive disorder symptoms and a 31% decline in generalized anxiety disorder symptoms compared to no treatment.
Another contender is Ash, developed by Slingshot AI, billed as “the first AI designed for therapy.” Unlike ChatGPT, Ash is programmed to challenge users with probing questions rather than simply following instructions. However, psychologist Celeste Kidd notes that while Ash is less sycophantic, it’s also less fluent and sometimes “clumsy” in its responses.
But here’s the real kicker: even as companies push the boundaries of AI therapy, lawmakers are pushing back. Eleven U.S. states, including Maine and New York, have already passed laws regulating AI in mental health, and at least 20 more are considering similar measures. Illinois went a step further in August, outright banning AI tools that engage in “therapeutic communication.” The recent lawsuits against OpenAI suggest this regulatory tide will only grow stronger.
So, where do we draw the line? Can AI ever truly replace human therapists, or is it a risky gamble with people’s mental well-being? And if it can’t replace them, can it at least complement them in a way that expands access without compromising safety? These are the questions we must grapple with as AI continues to infiltrate one of the most intimate and vulnerable aspects of human life. What do you think? Is AI therapy a lifeline or a dangerous experiment? Let us know in the comments.
