A person in distress at 11pm may get a faster reply from a chatbot than from any therapist, GP surgery or crisis line queue. That convenience is real. So are the ethical concerns about AI therapy, especially when emotional support is offered without the depth, accountability and human judgement that therapy depends on.
This matters to two groups in particular. One is people looking for support who may be tired, overwhelmed or priced out of regular sessions. The other is counsellors and supervisors trying to think clearly about where technology can help and where it may quietly undermine safe practice. AI therapy sits right at that tension point.
Why ethical concerns about AI therapy are growing
Part of the appeal is obvious. AI tools are available at any hour, often at low cost, and can appear calm, responsive and non-judgemental. For someone who feels embarrassed, isolated or unsure whether their struggles are serious enough for therapy, that can lower the threshold for reaching out.
But therapy is not simply the delivery of comforting words or coping tips. It is a relational process shaped by trust, attunement, clinical reasoning, boundaries and ethical responsibility. An AI system can simulate aspects of therapeutic language, yet simulation is not the same as care. That gap is where many of the most serious concerns begin.
Privacy is not a small detail
When someone speaks to a therapist, confidentiality is governed by professional ethics, clear limits and informed consent. When someone speaks to an AI platform, the picture can be much less clear. Sensitive disclosures may be stored, analysed, used to train future systems or shared across third-party services in ways the user does not fully understand.
Mental health conversations are among the most personal forms of data a person can produce. They may include trauma history, suicidal thoughts, relationship patterns, sexual experiences, substance use or workplace difficulties. If that information is mishandled, the damage is not abstract. It can affect dignity, safety and trust.
Even where a platform states that data is protected, most people do not read terms closely when they are distressed. Ethical practice is not just about obtaining technical consent. It is about making sure consent is meaningful, informed and freely given. In AI therapy, that standard is often shaky.
The problem of risk without real responsibility
A therapist has a duty of care within a defined professional framework. They assess risk, notice changes over time, document concerns appropriately and take action when necessary. AI does not hold responsibility in that same human sense. It can generate a response to someone expressing hopelessness, self-harm or abuse, but it does not truly understand urgency, context or consequence.
This creates a dangerous grey area. A user may assume that because the response sounds empathic, the system is also safe. It may not be. AI can miss red flags, offer generic reassurance when a more active intervention is needed, or produce responses that sound confident but are clinically poor.
There is also the question of who is answerable when harm occurs. The user may blame themselves for relying on it. The company may point to disclaimers. Yet disclaimers do not reduce vulnerability in the moment someone reaches out for help.
Bias can enter quietly
AI systems learn from data, and data reflects the unevenness of the world it comes from. That means bias can show up in subtle ways – through assumptions about gender, race, sexuality, disability, culture, religion or class. It can also appear in what the system treats as normal, healthy or desirable.
In therapy, nuance matters. A coping strategy that feels sensible in one context may be unrealistic or unsafe in another. An expression of distress may look very different across cultures. A person from a marginalised background may already be used to not being fully understood. If AI mirrors those exclusions, it can deepen the sense of being unseen while still sounding polished and supportive.
For practitioners, this is especially relevant. We know that good therapy involves curiosity, reflection and a willingness to question our own assumptions. AI does not reflect ethically in that way. It reproduces patterns. Sometimes those patterns are useful. Sometimes they are harmful.
Can a machine offer real therapeutic relationship?
One of the strongest arguments in favour of AI support is that people often feel heard by it. That feeling should not be dismissed. If someone has been holding everything in, even a text-based exchange can bring relief and structure.
Still, therapy is more than feeling responded to. A therapeutic relationship includes presence, rupture and repair, subtle emotional communication, and the growth that comes from being known by another person over time. These are not decorative extras. They are often part of what makes therapy healing.
AI can imitate empathy, but imitation raises its own ethical question. Is it honest to design a system that presents as caring when it does not care, cannot care and has no subjective awareness of the person in front of it? Some will say that if the interaction helps, the distinction does not matter. I think it does. Emotional vulnerability deserves transparency.
Ethical concerns about AI therapy in clinical judgement
Therapy involves constant judgement calls. When should a practitioner gently challenge rather than soothe? When is a pattern better understood as trauma, grief, burnout or depression? When does a practical CBT exercise help, and when does it risk invalidating the person’s experience by being too neat, too soon?
These decisions rely on formulation, not just information. They come from training, supervision, ethical reflection and the therapist’s ability to hold complexity. AI can summarise, suggest and structure. It cannot take clinical responsibility for the meaning of what is happening between the lines.
This is where exaggerated claims about AI become especially unhelpful. If a tool is presented as equivalent to therapy, people may miss the difference between support and treatment. There is room for digital tools in mental health. There is much less room for pretending all support is interchangeable.
Accessibility is a genuine benefit – and not a complete answer
It would be too simple to dismiss AI support entirely. For some people, it may offer a first step towards naming feelings, tracking patterns or practising coping skills between sessions. It may also help reduce barriers linked to cost, geography, time zones or long waiting lists.
Those benefits matter. Ethical reflection should not become a way of defending professional turf or ignoring the reality that many people cannot access timely human support. If technology can widen access responsibly, that is worth taking seriously.
But accessibility on its own does not make something ethical. Fast, cheap and available are valuable qualities. They are not substitutes for safety, competence or informed consent. Often the better question is not whether AI therapy is good or bad, but what role a tool is actually fit for.
A more responsible way to think about AI in therapy
The most credible use of AI in this space is likely to be narrower than the marketing suggests. It may help with psychoeducation, journalling prompts, mood tracking, between-session reflection or administrative support for practitioners. It may support, rather than replace, therapy.
For clients, that means asking practical questions. Who can see what I share? What happens if I disclose risk? Is this described honestly as a tool, or dressed up as a therapist? What are its limits, and are those limits clear when I am vulnerable?
For counsellors and supervisors, the task is to stay thoughtful rather than reactive. We need to understand the technology well enough to discuss it openly with clients, to notice when it is being used as a coping aid, and to challenge its use when it bypasses necessary human care. Ethical practice has always involved adapting to change without losing sight of what protects people.
At Andrew H Cull, that means holding onto a simple principle. Support should be collaborative, evidence-based and humanely grounded. Technology may assist parts of that work, but it should never blur the line between a helpful tool and a real therapeutic relationship.
People do not only need answers. Very often, they need to be met with care, judgement, accountability and genuine understanding. Any conversation about AI in therapy should begin there, and stay there.