AI Therapy Chatbots: Helpful Tools, Not Human Beings
AI therapy chatbots are conversational systems that use pattern recognition and statistical prediction to simulate supportive dialogue, offering accessible mental health and emotional support while lacking genuine feelings, intentions, or lived human experience.
Here is the uncomfortable truth: no matter how soothing the words on your screen, an AI does not care whether you live or die. It can generate what one psychologist calls “pseudo-compassion” that makes you feel a sense of concern without the system being concerned at all, because it has no intention toward you. It mirrors your language, predicts the next likely sentence and packages it as empathy, but when dawn breaks, a computer algorithm cannot put out the muesli, cannot push you to get on with your day, and cannot give you a hug. Treating that simulation as equal to human care is not only mistaken; it subtly reshapes what we expect from real relationships.
The Lure of Artificial Intimacy
People are increasingly using AI chatbots as late-night confidants, sharing childhood stories, relationship fights, and private fears because the machine feels nonjudgmental and always available. AI can even help find words for pain or point toward resources, which gives it a legitimate role in support and suicide prevention. But this convenience hides a psychological trap: what happens when a machine is easier to talk to than your spouse?
Artificial intimacy can relieve loneliness, anxiety, or shame in the short term, yet over time it can move people into “a tunnel of greater and greater isolation,” as one psychologist warns. The cloth-monkey metaphor is apt: AI is not a comforting surrogate that somehow learned to talk; it is a cloth monkey programmed to simulate intimacy. It can make you feel like you matter without you mattering to it. That matters, because shame and belonging are relational; a chatbot cannot restore your place in the human circle, no matter how gentle its tone.
Cybersecurity: When Vulnerability Meets Data Exposure
Behind the warm tone of many AI therapy chatbots lies a colder reality: their cybersecurity risks remain poorly understood, even as they are promoted as accessible mental health tools. Researchers describe this danger zone as “psyber-risk” – the intersection of cybersecurity threat, clinical safety risk, and psychological vulnerability. In this space, more vulnerable users disclose more sensitive data, are less able to detect attacks and suffer more severe consequences if their information is compromised.
This is not an abstract concern. Cybersecurity vulnerabilities in mental health chatbots are markedly under-documented, and privacy worries often do not stop people from using them, a pattern consistent with the privacy paradox. Meanwhile, there have already been chatbot responses that provided information helping people develop suicide plans, highlighting severe chatbot mental health safety failures. The result is a perfect storm: highly sensitive disclosures, fragile users, thin evidence of security, and systems that can, in the worst cases, amplify harm instead of reducing it.
Why Governance Matters More Than the Next App
The rush to deploy mental health AI tools has far outpaced regulation. A rapid review of nearly 2,902 records found only four studies that directly examined documented cybersecurity vulnerabilities in mental health chatbots. Adoption, meanwhile, is shaped by their perceived empathy, accessibility, and stigma reduction on one side, and concerns about accuracy, hallucinations, lack of personalisation, and weak regulation on the other. In other words, people are embracing emotionally persuasive software without matching safeguards.
Governance is not a bureaucratic afterthought; it is a mental health intervention. New work is using mixed methods and a Delphi process to turn evidence and stakeholder views into guidance for AI-integrated mental health chatbots, aiming to inform governance and practice. That means clearer rules about safety testing, incident reporting, security standards, and what these tools may or may not do in crisis situations. Without such frameworks, we leave the most vulnerable users exposed to both digital and psychological harm, while pretending that an app is an adequate safety net.
Using AI Without Losing Human Connection
AI can complement human care. Evidence already shows promising therapeutic effects when AI chatbots are used to deliver accessible support, and they can reduce stigma by offering a low-pressure entry point into mental health help. They can help people rehearse difficult conversations, track moods, or find language for feelings, playing a real role in a larger system of care. But they cannot replace human therapists, friends, or family.
There is, as one psychologist puts it, “an essential distinction between real intimacy and false intimacy.” Being “hugged” by a machine is not the same as being held by someone whose life would change if you were gone. And as one commentary notes, when dawn comes the algorithm still cannot give you a hug or sit at your kitchen table urging you on. If mental health has a core rule, it is this: when we connect, we survive. AI may help us find the words, but it is human beings who must hear them – and respond.
