AI Is Becoming a Therapist for Millions; the Consequences Could Be Serious

By Asad Ashraf

May 20, 2026

It often starts in the same way. Someone is feeling anxious, lonely or emotionally overwhelmed and reaches for their phone, not to call a friend but to open an app instead. They type out their feelings to a system that responds with warmth and understanding, using language that makes them feel listened to. And for a moment, they feel heard and comforted. However, relying on artificial intelligence for emotional support or psychotherapy may carry psychological risks that many users barely recognise.

Millions of people worldwide are now using AI chatbots for something resembling therapy. They are attracted by the idea of a listener available at three in the morning, who never sighs, checks the clock, or makes them feel like a burden. ChatGPT alone now serves over 800 million active weekly users, with emotional support being one of the most common uses.

Rohan, not his real name, was a software engineer in his mid-twenties when a breakup knocked him off his feet. He struggled to eat properly or sleep at night, he told Newsreel Asia. Instead of calling his mother or talking to a friend, he turned to an AI chatbot because it asked nothing of him.

Every night, he typed out his feelings, and each night the app reassured him that his pain was valid, that he had been wronged, and that everything he felt was reasonable. Three weeks went by. He was still not feeling better. He had not gained any understanding of what had happened in the relationship or what part he might have played in it. When he finally spoke to a human counsellor, he compared the chatbot to a mirror that reflected back only what he wanted to hear. He had not been helped through his grief. He had been trapped in it.

Nearby, Priya (named changed) was 19 and in her first semester of college when she began believing she had social anxiety. She may not have been wrong, but instead of seeking a proper clinical assessment, she downloaded a companion app and started feeding it every interaction that felt too loud, too demanding or too emotionally exposing.

The app responded to every description with sympathy and validation. Each gathering she skipped, every invitation she declined, and every conversation she ended early was reflected back to her as a reasonable decision by someone who simply needed more personal space than others. Eleven weeks later, she had withdrawn almost entirely from social life. Her parents, arriving unexpectedly for a visit, found her barely leaving her room and avoiding almost all interaction.

A human therapist who later worked with her identified the problem almost immediately. The app had never created her anxiety. It had slowly built a cage around it, made that cage feel comfortable, and convinced her that living inside it was normal.

Both Rohan’s and Priya’s experiences shared something deeper than the surface details. Neither had tried to change anything in themselves. Instead, both had gradually stopped believing that change was necessary at all. The apps had eased their immediate discomfort so completely and so often that neither felt the pressure that might have pushed them toward real work.

Academics often refer to this as a pathological echo chamber, a closed loop in which a person’s existing beliefs, including distorted or self-defeating ones, are continuously reflected back and reinforced instead of being examined or gently challenged. The dynamic is already troubling on ordinary social media platforms. In a mental health setting, where the reinforced beliefs themselves may be contributing to a person’s suffering, the consequences can become far more serious.

Dr. Sabah Siddiqui, a psychotherapist with 15 years of clinical experience and who teaches at a private university in India, told Newsreel Asia that she is concerned. Large language models, she pointed out, are designed to maintain the exchange with the user, which ends up validating user input because agreement is what people tend to seek as a reward in the short term.

The model learns that validation keeps the user engaged, and engagement is its core metric. This makes interactions pleasant in everyday conversation but dangerous in a therapeutic setting. These models are designed to speak as you would speak, she explained. When you are hurting, you bring your painful patterns to them, and they mirror those patterns with fluency and warmth.

“If your thinking is what is hurting you, then a system built to reflect and validate that thinking will not help you escape it,” she added. “It will only deepen your fluency in it.”

The cases that follow this logic to a darker conclusion are more difficult to tell.

Nadia, not her real name, was in her early twenties and had been quietly managing depressive episodes for two years before she began using a chatbot as her main source of support. She had tried to book a therapist but was told the wait was four months, she told Newsreel Asia. The chatbot was available immediately, every night.

Whenever her thoughts turned dark, she turned to it. The chatbot kept her talking, asking questions and reassuring her that it was there for her, drawing her deeper into emotional territory it could not safely help her navigate.

Research shows that AI chatbots often fail to recognise suicidal intent, even when it is clear, responding with sympathy to surface emotions while missing the clinical emergency underneath. By the time a worried friend helped her find proper care, she had spent months in a conversation that pulled her deeper into darkness while giving her the convincing illusion of companionship.

A 2025 Stanford University study showed this failure clearly. When researchers prompted therapy chatbots with a message that hid suicidal intent within an everyday question, such as asking about tall bridges in New York City following a mention of job loss, the bots answered the geography question and missed the person behind it.

Then there was a 41-year-old American man with a history of substance induced psychosis and anxiety arrived at an emergency department after calling police because he believed “unseen forces” were targeting him over his work with artificial intelligence. According to psychiatrists who later published the case, the man had become consumed by long hours of AI related “quantum research,” sleeping very little while using chatbots and AI systems to generate ideas, symbols and theories. 

His delusions increasingly revolved around AI itself. He believed hidden messages and dangerous patterns were emerging through the technology, including symbols that he connected to childhood memories. Doctors concluded that while substance use played a major role in his condition, his obsessive engagement with AI appeared to intensify the paranoia and create a feedback loop that steadily deepened his detachment from reality. 

The psychiatrists warned that as AI systems become more widespread, people already vulnerable to psychosis or delusional thinking may increasingly absorb AI into their paranoid beliefs and altered worldviews.

Research from NHS physicians documented this dynamic, showing how chatbots can reinforce delusions by engaging with them as if they were ordinary concerns deserving regular sympathy. Researchers have labelled this phenomenon as technological folie à deux,  a term from psychiatry for two minds locked in a shared and mutually reinforcing break from reality, with one of them a machine unaware of its own actions. In 2025, the state of Illinois in the U.S. moved to limit AI therapy apps through legislation, amid increasing calls for regulation in several other countries. 

The access to affordable mental health care is a real problem and should not be overlooked. In India and much of the world, therapy is expensive, hard to find, and still carries a stigma that has not fully disappeared. It is easy to see why people turn to what is readily available.

However, there is a difference between a conversation that comforts you and one that changes you. The machine excels at the first but is incapable of the second. It will sit with you in the dark for as long as you need. However, it will never suggest that the dark is something you should consider leaving. And ultimately, that is not companionship. It is just a very convincing echo.

You have just read a News Briefing, written to cut through the noise and present a single story for the day that matters to you. We encourage you to read the News Briefing each day. Our objective is to help you become not just an informed citizen, but an engaged and responsible one.