Understanding AI Phobia in Mental Health

By Eduardo Bunge, PHD

Before you insult me for insinuating that there is an “AI phobia,” hold a second and please read at least a couple of paragraphs. 

If you have intense feelings against AI, you will of course object to the very idea of AI phobia in mental health. After all, many concerns about AI are legitimate. A phobia is an excessive fear that leads us to avoid the feared stimuli and interferes with our functioning.  So what do I mean by AI phobia in mental health? I mean, an immediate reaction, characterized by excessive fear that shuts down the careful consideration of our fears about AI, impedes the possibility of using it with our clients and limits our functioning as clinicians. If I told you to embrace AI without any concerns, you will think I'm crazy. Similarly, if we reject AI completely, without evaluating our fears, the pros and cons and even the evidence, we would make a similar mistake.

I believe there is a healthy, responsible way of integrating AI into the field of mental health, but the amount of fear about AI does not allow us to challenge some of our concerns. As therapists, we help our patients examine their fears every day. We teach them to stop and think, before accepting them automatically.  We do not dismiss their fear, but intense emotions require intense examinations. We help our clients identify what is valid, what may be exaggerated, and what alternative perspectives might exist. If after careful examination, they determine the fear is valid then adjust to it, but if the fear is amplified we recommend testing those fears and actually face them. 

We should do the same with our fears about AI. At least we should ask ourselves “What are the risks of not adopting AI?” Why do this? Because we might be missing an opportunity. What if we are turning away from a tool that could help millions of people access support, learn skills, and receive care that they otherwise would never receive? So here’s my proposal. I would like us to understand: 1. why we should care about AI in mental health, 2. where our fears come from, and 3. examine each fear.

Quick note, I wrote this from multiple perspectives, I am a psychology professor, researcher, former clinician, consumer of therapy (with a human therapist!) and founder. If interested in my background read the note at the end.

Why should I care about AI in mental health

To understand why we should care about AI in mental health we need to acknowledge something uncomfortable: The current mental health system is failing to help millions of individuals. More than 1.2 billion people worldwide live with a mental disorder. Some estimates say that 80% never receive treatment (WHO, 2025). Those who do access care often face long waitlists, high costs, limited availability of trained professionals, and significant barriers to care. Even when evidence-based treatments are available, outcomes are far from perfect. Dropout rates remain high. Many people fail to improve and remission rates are often much lower than we would like. For example, anxiety disorders, one of the areas where psychotherapy performs best, generally yield a 51% remission rate and 64% response rate (Roy-Byrne et al., 2010). That means that for all other disorders we are below that! 

I am a therapist myself, I am simply acknowledging that our current treatments and the system are not meeting the needs of everyone who requires help. Even if you are convinced that we are doing well, we should constantly be searching for ways to improve it. We therapists did not cause the mental health system problems, but we can work on improving them (and we should).

I’d like to propose a brief thought experiment to picture the opportunity we might be missing. For a few minutes, let’s imagine that our concerns about AI have been solved or at least substantially reduced. I know, we are still not there yet, but let’s suspend that concern for a moment. Imagine that we had AI systems designed by mental health professionals, for mental health professionals. If these were true, this is what AI would be able to do:

• Support clients between sessions

• Reinforce therapeutic skills worked in the sessions

• Provide psychoeducation.

• Assist with assessments.

• Help clinicians identify important patterns.

• Support clients when therapists are unavailable.

• Improve access to care.

• Accelerate the pace of treatments

• Advice therapists when needed.

• And many more…

Would that be a good thing? Imagine how many more people might feel supported? How many clients will finally be able to leave less anxious lives, with less depression, with fewer psychotic symptoms, how many of our loved ones will get better care. I suspect most clinicians would answer “this might be good, but what about all the risks?” 

If we believe some of these benefits are possible, then it is worth analyzing our fears as we do with our own clients. So let’s understand where these fears come from.

Where does my AI fear comes from

I believe our collective perception of AI is being shaped by two phenomenons: Our tendency to attend to dangers more than opportunities and “the fear industry.” 

The history of the mental heald field shows that professionals have a tendency to resist innovation. We once resisted treatment manuals, evidence based paradigms, self help programs (even when some showed small positive effects). These resources were all criticized at some point for being impersonal, ineffective, or even harmful. Today, most are considered normal parts of care. More recently we resisted digital interventions (such as smartphone applications), and again, today almost no one has a problem with, and apps such as headspace for meditation, are widely used. But the most clear example of fear and resistance is what happened with Videoconference Therapy, we resisted it for almost 20 years, and only when the pandemic hit, we realized that we could actually help many clients. While we were debating those technologies, many people remained underserved, and that is a risk that we should be held accountable for. So there is risk in adopting AI but there are also risks in avoiding it, and we should not ignore the risks of inaction. Across 20 years of conference presentations on digital interventions and more recently on AI studies, I consistently get tons of questions about the potential risks. Despite presenting about different technologies, the questions are almost identical. So I started to wonder: are we witnessing a moral panic? 

Here is where the industry of fear comes to play. There is an entire ecosystem built around capturing our attention based on fears that propagate what is called moral panic. A moral panic occurs when a new technology, behavior, or social change becomes viewed primarily through the lens of its dangers, often leading society to overestimate risks while underestimating potential benefits. Stories about harm generate more clicks than stories about benefit. A teenager harmed by technology becomes a headline. Millions of individuals helped by technology rarely does. As a result, we are constantly exposed to what can go wrong and rarely exposed to what can go right. This does not mean the risks are not real. It means we may be seeing only part of the picture. Furthermore, the risks of the traditional models of mental health are not a click bite at all. If someone suicides at a clinic (which sadly happens very frequently), that does not make the headlines, and it does not meant the clinic should be shut down. 

Just as an example, there are 800 million users of Open AI alone. Of course we will read tragic news related to this new technology, but how representative is that of the benefits-risk ratio? If there are 799.999 users that are just ok or even feeling better, we will hear nothing about them. So we are constantly being exposed to biased media. Of course a single case might mean everything to that individual and efforts need to be done to prevent any tragedy (see "dangerous advice” section). With this in mind, let’s start examining the most common fears associated with AI.

Most frequent AI fears in mental health

Here comes a list of fears I frequently see in the field. Hint: I share some, but I disagree with several of them. Next, I will share what supports those concerns, what doesn’t and what are the opportunities we might be missing. 

1. AI is bad.

2. All AIs are the same.

3. AI will replace therapists.

4. There is no Privacy/confidentiality.

5. The field is largely unregulated.

6. Technology companies only care about money.

7. AI gives dangerous advice and increases suicide risk.

8. AI is just psychofancy.

9. With AI they only care about reducing costs

10. "AI will expand and will take over the world”

"I hate AI, AI is bad."

Valid concern: AI can absolutely be used in harmful ways. It can spread misinformation, reinforce unhealthy behaviors, and create unintended consequences. These risks should not be ignored.

Potential exaggeration: Saying "AI is bad" is like saying "the internet is bad." The statement is too broad to be useful. Using the internet analogy, we now know that it has created enormous problems but the benefits exceed them by far. As therapists, we often challenge generalizations and all-or-nothing thinking in our patients and we should challenge it in ourselves.

Alternative thought: One of the greatest achievements of the internet was democratizing access to information. Similarly, AI may help democratize access to support, education, coaching, and evidence-based mental health resources. If millions of people continue using these tools voluntarily, "What value are they finding in AI?" Do they use it only because the AI is agreeable or are they getting something that they can’t get somewhere else? Of course we would also need to understand, who is getting real value? How? When? And who should not be using it?

"All AIs are the same."

Valid concern: The most known AI systems are designed for general purposes, and poorly designed for mental health. Not all AIs should be trusted as a valid psychological support. Character AI, is an entertainment tool, not an AI for mental health, and I would not encourage a client to use Character AI for mental health support. 

Potential exaggeration: We often talk about AI as if it were a single thing. In reality, AI is a category. There is a major difference between general-purpose systems such as ChatGPT, Claude, and Gemini and mental-health-specific systems such as Therabot, Ash, Wysa, Parente, and others. Mental-health-specific systems are designed around therapeutic outcomes rather than general engagement. They include clinical safeguards, supervision, monitoring, and provide evidence-based interventions. Rejecting all AI may cause us to overlook systems specifically designed to improve mental health outcomes and that could support the clients and even the therapists!

After thought: The mental health AIs mentioned above have been publishing research showing promising evidence. Did you try them? I suggest those that oppose AI to give them a try, they might be surprised. 

"AI will replace me."

Valid concern: AI will likely change the profession. Some tasks may become partially automated. Some mild concerns may increasingly be addressed by AI.

Potential exaggeration: The idea that therapists will disappear assumes that AI and therapists are direct competitors. In reality, mental health needs vastly exceed the capacity of the existing workforce. There are more than 1. billion people living with mental disorders worldwide and 91% of those with depression do not access care! (WHO, 2025). Many receive no support at all and even evidence based interventions help a portion of the clients, and we still have severe mental conditions that require intensive support.

Opportunity we may be missing: Some people will prefer humans. Some will prefer AI. Many will want both. A therapist using AI responsibly can provide support beyond the therapy hour. Our future work won’t be us versus AI but rather therapists and AI working together. Think of AI as computers. Therapists without computers have fewer clients than those with computers, but more importantly, therapists with computers can help remote clients that otherwise they would have never got support. Therapists with AI will probably have more work (and better quality) than therapists without AI. I can bet money on this :)

"Everything I share with AI can be used against me and my clients."

Valid concern: Privacy and confidentiality are legitimate concerns. Especially in those AIs that are not designed for mental health. Mental health information deserves the highest level of protection.

Potential exaggeration: Not all AI systems manage data in the same way. Healthcare-specific systems often include safeguards, privacy protections, and HIPAA-compliant processes. Many technologies initially triggered similar fears. Online banking and online payments once seemed frightening. Today, most people use them safely because the benefits outweigh the risks when appropriate safeguards exist. Although there is no zero risk and some people get scammed, the goal is responsible risk management.

Alternative thought: Users can be educated about this and can decide what they disclose or not. For example, we can instruct clients to use aliases, and not share anything that they consider extremely valuable for them. It is not the same to share “My child is throwing a tantrum” than sharing experiences of being abused.

“The field is largely unregulated.”

Valid concern: This is one of the concerns with which I have the greatest sympathy. Artificial intelligence is an extraordinarily powerful technology. And like any powerful technology, it can be used responsibly or irresponsibly. The regulatory landscape remains uncertain in many jurisdictions. Some governments have moved toward stricter restrictions, while others have taken a more permissive approach. In many places, significant gray areas remain. Questions about accountability, transparency, safety, privacy, clinical responsibility, and professional oversight are still being actively debated. I believe these concerns are legitimate.

Alternative thoughts: The existence of regulatory uncertainty does not automatically imply that every use of AI is unsafe. Many technologies are introduced before regulatory frameworks fully mature. The appropriate response is not necessarily prohibition, it is careful implementation, ongoing evaluation, and thoughtful regulation. Treating every AI application as equally risky may prevent us from distinguishing between responsible and irresponsible uses.

An important distinction: AI can be used as a standalone resource or integrated to a therapist. To me, this distinction is important and regulations should keep this in mind. An AI system operating entirely on its own is a very different use case from an AI system functioning under the supervision of a trained mental health professional. The future of mental health AI may not be found at either extreme. It may be found in carefully designed models where human expertise and artificial intelligence work together, each compensating for the limitations of the other.

An AI operating under professional supervision is one of the most promising applications of AI. This is the approach that has guided much of my own work. Rather than replacing clinicians, AI can support clinicians, can function within systems that include monitoring, review, accountability, and professional judgment. In these models, AI becomes an extension of care rather than a substitute for care.

"Technology companies are bad and only care about my (and my clients’) money."

Valid concern: Technology companies have financial incentives.Some have made mistakes.Oversight and accountability are important.

Alternative interpretation: The existence of profit does not automatically imply harm. The pharmaceutical industry, despite its flaws, has helped save countless lives. Economic incentives exist throughout healthcare, including among therapists, hospitals, insurance companies, universities, and private practices. In fact the fear about losing our jobs reflects our (valid) financial concerns. 

The most important metric should be outcomes. Not intentions. Financial incentives need to be controlled, but whether we like it or not, they affect all of us. Even the therapists in their private practices or at non-for profit agencies, we all still need to pay our bills. Ultimately, if a tech company reduces mental suffering that should be good news, for all those that came into this field with the mission of reducing human suffering.

"What if an AI gives dangerous advice to my clients?"

Valid concern: There have been tragic cases. These should be taken seriously. Researchers, clinicians, and developers should work continuously to improve safety. There reports of users of general AI’s that suicide, increased psychotic symptoms, increased obsessions, etc…actually the list is long.

Potential exaggeration: News reports focus on tragedies. We need to look at a large data set and run empirical studies. The news rarely report ordinary successes. So we may be seeing a distorted sample of outcomes. We also tend to apply different standards to AI than we do to human clinicians. Human therapists miss suicide risk. Human therapists make mistakes. Human therapists sometimes fail to help.

Compromise: When a person is struggling at 2 a.m., the comparison is often not between AI and an available therapist. The comparison is between AI and no support at all. The question should not be whether AI is perfect. The question should be whether it can help more people than the alternatives currently available. Of course we need to conduct studies to compare the risks and the benefits, but let’s remember that human therapists alone have a hard time battling suicide and many more other mental health conditions. The risks exist on all sides, AI alone, Humans alone and even in the combination of AI and human therapists, the goal is to reduce those risks as much as possible. Until proper RCTs are conducted we can not confirm or deny one or the other, we can only speculate. So far the only proper RCT conducted has shown greater improvement in depression and anxiety and body and shape concerns in college students using Therabot, an AI as a standalone resource (see Heinz, et al.,  2025).

"AI is just psychofancy."

Valid concern: AI can sometimes reinforce beliefs too readily. This can be particularly concerning for vulnerable populations, including individuals experiencing psychosis or obsessions. May users do not like that the AI always tell them “great idea” or “that is a valid concern”

Alternative views: Therapists are trained to validate our clients and it is really hard for therapists to know when we are validating enough or in excess. Actually, the tendency to agree with people is not unique to AI. Friends do it. Family members do it. Therapists do it. Validation is not bad in essence, and all humans are vulnerable to confirmation bias. Many clinicians have experienced moments when they later realized they had accepted a patient's narrative too quickly. AI ‘s designed for mental health purposes aim to encourage perspective-taking, curiosity, and critical thinking. Whether they are doing it appropriately or not, will only be answered with more studies. Interestingly, there are some studies where human judges rated AI responses as more empathic than human responses. So, let’s remind ourselves that this is not an easy answer. Anecdotally, most of the Parente users tend to like that the AI validates them, however there are some that do not like it at all. Finding that nuance level of how much validation for which clients is not easy for humans and Ai.

"AI proponents are only interested in reducing costs"

Valid point: reducing costs can come with a reduced quality, and that is a valid concern. Especially when therapists don’t get paid appropriately. 

Counter argument: Therapy is expensive for most individuals in the world. If the cost is lower, that means more people will be able to access it. If therapists start working with AIs it will probably require a reshuffling of the economic side of the mental health system. An interesting study showed that combining AI plus human therapists achieved comparable outcomes to pure human care while reducing clinician time by up to 8 times! (Palmer, 2025). So the quality remained the same. We had some promising experiences (at Parente) delivering groups led by therapists, supported by Ai, with positive outcomes and without increasing the burden to the therapists and improving their compensation! There are several business models that will require careful consideration. But here is my take, if by using AI, we can help more clients, without sacrificing quality, compensation should be higher, and the system will probably still work.

"AI will expand and will take over the world”

Valid point: Yes, one reality we need to acknowledge is that AI will continue to evolve whether we like it or not and no one has the crystal ball. Uncertainty is typically a source of stress for many of us.

Counter argument: technology has evolved across the history of humanity, there were always these type of fears, and those defending the new technologies got the same reaction from the skeptics “but this time is different.” However, we live in a world with more comfort and safety than ever before. So far we managed to use technologies to our favor more than against us.

My take: We can choose to remain on the sidelines criticizing it, or we can become actively involved in understanding it, studying it, regulating it, and helping ensure that it develops in ways that benefit patients. Because if AI is going to become part of society regardless, I would rather have psychologists, psychiatrists, researchers, and patients helping shape its future than leaving those decisions entirely to technologists.

"I don’t like AI, I don’t want to use it myself nor my clients”

Valid point: if you don’t like it, AI is of course not for you. That is completely respectable. This is not a fear but worth thinking about it, it is a”preference” and that’s okay. No one is expecting that everyone uses AI.

Counter argument: Not everyone has the same preferences. Some people will want: Human only therapy, others, AI only due to the stigma or embarrassment of talking to another human, and others will want to have a therapist and AI for those moments in which the therapist is not available. Actually, patients are already choosing AI, 1 in 5 young adults use chatbots for mental health (McBain., et al. 2025)), comparable to those in professional therapy and 93% find it helpful. Whether we like it or not, it's happening. As long as the client is a consenting adult that knows if they are chatting with an AI or not, they are the ones that should make the choice, not us. What is important for us to understand is who benefits from the AI, how, and under what conditions.

Conclusion

Every major technology has generated fear. Some fears turn out to be justified. Others turn out to be exaggerated. As clinicians, we know that fear should be explored rather than blindly accepted. I do not propose to blindly adopt AI, I propose thoughtful evaluation, responsible implementation, and continuous improvement. By analyzing each of the most common fears I have heard in my colleagues, I hope this helps move the discussion forward and that we prevent gut reactions to feared technologies. We need to ask ourselves "What happens if we use AI? AND What happens if we don't?" And as with any phobia a gradual exposure may allow us to examine our fears better. As a first step we should test it in a controlled environment as the research field, and many researchers have done it. Let’s please check the evidence, there are meta analysis even with the old AI (the rule based chatbots) that show small improvements and the newer publications based on generative AI show promising outcomes too (He et al., 2023). A potential next step in our gradual exposure could be to use AI under the monitoring of a therapist, and only open it when we know the chances of benefiting the clients exceed the risks. But avoiding the AI discussion and or implementation, just because of the fears, looks more like a phobia than the curious or scientist approach we suggest to our anxious clients.

About the Author / Where I'm Coming From

As I mentioned, I am a psychology professor, researcher, licensed therapist in Argentina (MN 33012), consumer of therapy (with a human therapist!) and founder. If interested in my background read the note at the end.

As a psychology professor who has spent more than 25 years studying, teaching, and supervising evidence-based psychotherapies, so I am all about evidence-based interventions, not just opinions. As a researcher I dedicated my career to studying digital mental health interventions and, more specifically, artificial intelligence since 2019. As a clinician I spent years seeing patients and families every day, I deeply value the therapeutic relationship, and encounter tons of challenges in making treatments effective for the families I was trying to help.

As a regular human being, I am currently seeing a therapist for personal matters but I also use AI for reflection, support, and guidance between sessions. These experiences have made me aware of several benefits of AI that are often overlooked, dismissed, or simply absent from the public conversation.

I want to share something personal. Back in 2019, during one of our earliest studies, I was reading participant feedback when I came across a comment that stopped me in my tracks:"I thank God for putting you in my path." The message was directed to a rule-based chatbot. Those annoying ones that we used to see in customer services. Even with that rudimentary technology many users were sharing similar comments. Clearly, the AI agents were mimicking something very human “conversations” and users were humanizing them. I had no doubt there was a real opportunity here, if used correctly, so I decided to focus all my research efforts in AI. Study after study, I got more convinced of the potential benefits of AI to the point that I decided to approach my work outside academia, become a founder and build a product that could be used by therapists working with parents of children with behavioral and emotional challenges.

References

Bunge, E. L., & Desage, C. (2025). A framework for evaluating mental health artificial intelligence-based conversational agents. Journal of Technology in Behavioral Science. https://doi.org/10.1007/s41347-025-00519-w

Desage, C., Patterson, T. P., Wu, Y., Bechtel, A., Bharat, A., Vaclavik, D., & Bunge, E. L. (2025). Illustrating a framework for assessing generative artificial intelligence-based conversational agents for mental health. SSRN. https://doi.org/10.2139/ssrn.5245727

Escoredo, M. C., Mostovoy, K., Schickler, R., Bechtel, A., Shagan, J., & Bunge, E. L. (2025). Enhancing parental skills through artificial intelligence-based conversational agents: The PAT Initiative. Family Relations, 74(3), 1250–1265. https://doi.org/10.1111/fare.13158

Fitzpatrick, K. K., Darcy, A., & Vierhile, M. (2017). Delivering cognitive behavior therapy to young adults with symptoms of depression and anxiety using a fully automated conversational agent (Woebot): A randomized controlled trial. JMIR Mental Health, 4(2), Article e19. https://doi.org/10.2196/mental.7785

Heinz, M. V., Mackin, D. M., Trudeau, B. M., Bhattacharya, S., Hekler, E. B., & Jacobson, N. C. (2025). Randomized trial of a generative AI chatbot for mental health treatment. NEJM AI, 2(4), Article AIoa2400802. https://doi.org/10.1056/AIoa2400802

Li, H., Zhang, R., Lee, Y.-C., Kraut, R. E., & Mohr, D. C. (2023). Systematic review and meta-analysis of AI-based conversational agents for promoting mental health and well-being. npj Digital Medicine, 6, Article 236. https://doi.org/10.1038/s41746-023-00979-5

MacNeill, A. L., Doucet, S., & Luke, A. (2024). Effectiveness of a mental health chatbot for people with chronic diseases: Randomized controlled trial. JMIR Formative Research, 8, Article e50025. https://doi.org/10.2196/50025

McBain, R. K., Bozick, R., Diliberti, M., Zhang, L. A., Zhang, F., Burnett, A., Kofner, A., Rader, B., Breslau, J., Stein, B. D., Mehrotra, A., Uscher-Pines, L., Cantor, J., & Yu, H. (2025). Use of generative AI for mental health advice among US adolescents and young adults. JAMA Network Open, 8(11), Article e2542281. https://doi.org/10.1001/jamanetworkopen.2025.42281

Palmer, C. E., Marshall, E., Millgate, E., Warren, G., Ewbank, M., Cooper, E., Lawes, S., Smith, A., Hutchins-Joss, C., Young, J., Bouazzaoui, M., Margoum, M., Healey, S., Marshall, L., Mehew, S., Cummins, R., Tablan, V., Catarino, A., Welchman, A. E., & Blackwell, A. D. (2025). Combining artificial intelligence and human support in mental health: Digital intervention with comparable effectiveness to human-delivered care. Journal of Medical Internet Research, 27, Article e69351. https://doi.org/10.2196/69351

Rivera-Cepeda, C. F., Pineda, B. S., Vaclavik, D., Bagner, D. M., Hardan, A. Y., Abadi, A., Mostovoy, K., & Bunge, E. L. (2026). Real-world data from a group parent management training program enhanced using artificial intelligence: Qualitative study. JMIR Formative Research, 10, Article e91841. https://doi.org/10.2196/91841

Rivera-Cepeda, C. F., Vaclavik, D., Bagner, D. M., Hardan, A. Y., & Bunge, E. L. (2025). Feasibility, usability, and promise of a parent management training using a generative artificial intelligence platform. Evidence-Based Practice in Child and Adolescent Mental Health. Advance online publication. https://doi.org/10.1080/23794925.2025.2602466

Roy-Byrne, P., Craske, M. G., Sullivan, G., Rose, R. D., Edlund, M. J., Lang, A. J., Bystritsky, A., Welch, S. S., Chavira, D. A., Golinelli, D., Campbell-Sills, L., Sherbourne, C. D., & Stein, M. B. (2010). Delivery of evidence-based treatment for multiple anxiety disorders in primary care: A randomized controlled trial. JAMA, 303(19), 1921–1928. https://doi.org/10.1001/jama.2010.608

Sohn, J.-S., Ha, B.-G., Park, S., Kim, J.-J., Lee, E., Oh, H., Lee, S., & Kim, E. (2026). Systematic review and meta analysis of chatbots in the management of depressive and anxiety symptoms. npj Digital Medicine, 9, Article 377. https://doi.org/10.1038/s41746-026-02566-w

Vaclavik, D., Desage, C., Bunge, E. L., & Bagner, D. M. (manuscript submitted for publication). Feasibility of an AI platform to augment parent management training: A case series.

World Health Organization. (2025). World mental health today: Latest data. https://www.who.int/publications/i/item/9789240113817

---

Dr. Eduardo Bunge is co-founder and CEO of Parente, a psychology professor at Palo Alto University, and Director of the Child and Adolescent Psychotherapy and Technology Research Lab. He has spent over 20 years researching and delivering evidence-based interventions for children, adolescents, and families, with a particular focus on Spanish-speaking communities.

Connect with Eduardo on LinkedIn or reach out directly at ebunge@parente.ai