Welcome To The Wild West: My Fears And Hopes About AI as a Trauma Therapist
Laurel van der Toorn, LMFT
Laurel is a trauma therapist licensed in multiple states. She takes a balanced position on AI and mental health; there will be good, bad, and ugly. She does not currently use or plan to use any AI in client-facing parts of her practice.
The most recent season of The Gilded Age on HBO felt especially resonant. It’s been about six months since it aired, but I keep thinking about it. While most of us can't relate to the specific tribulations of high society in 1880s Manhattan, the show's depiction of rapid industrial acceleration and widening wealth disparities feels remarkably parallel to today. As we watch the character based on Cornelius Vanderbilt stop at nothing to extend the railroad across the US, we see AI rapidly expanding into the world around us.
Just as Vanderbilt's railroad pushed into the Wild West to accelerate trade, AI is in its own Wild West era today. Few regulations exist to structure and contain how AI can be used. Local and federal governments are scrambling to regulate an industry unlike any that came before it. The regulations that do exist often contradict those of other districts, making it impossible for AI companies to remain fully compliant. Meanwhile, billions of dollars are being poured into AI development, pushing the industry forward at lightning speed.
In short, it’s messy.
Many of us are grappling with ethical concerns about AI in general. Unclear privacy policies, environmental impact, and the potential elimination of many professions are concerning issues to consider.
But regardless of where our ethics fall, AI is here to stay. This reality brings with it a great deal of anxiety and fear; we all want job security, and most of us don't love change. Many of the high-achieving professionals I work with in therapy (lawyers, founders, physicians, and tech leaders) are grappling with this question right now: where do humans still matter?
We’ve seen a version of this before in mental health. Teletherapy was once considered controversial too. Many therapists worried it would dilute the therapeutic relationship or reduce the depth of the work. Instead, teletherapy expanded access to care and allowed people to work with therapists who truly specialized in their needs. Today it’s a normal, effective part of mental healthcare. There are some therapists and clients who still prefer to work in person. To each their own; no one is doing any harm with their stance. AI may follow a similar path; not replacing therapy, but reshaping how people access it.
As a therapist, I see plenty of reasons to be hopeful about how AI will reshape mental healthcare. And of course, there are plenty of reasons to be wary. Below are my thoughts and things I'll be monitoring as this technology matures.
Reasons I'm Hopeful About AI As A Therapist
Just like Vanderbilt's railroad, I'm cautiously hopeful about how AI will make therapeutic material and information more accessible to people who can't or won't go to therapy. This is a good thing. Information is power. Just as Google didn't destroy education or eliminate the need for schools, I don't think AI will destroy therapy either.
I often see people arrive in trauma therapy after years of trying to solve their struggles alone with books, podcasts, or internet advice. If AI helps people get closer to the right language for what they're experiencing, that could be a meaningful first step.
Early findings are pretty clear: AI does some things better than humans when it comes to therapy. It's great at consistently showing empathy and offering encouragement. It's also proven to be quite effective in treating uncomplicated anxiety and depression.
While AI can't calm someone the way sitting in the presence of a therapist who has a grounded nervous system does, it can definitely offer great coping strategies. It can suggest breathing exercises, healthy distraction techniques, and mental reframing. It's available 24-7 and can stay emotionally neutral in a way most humans cannot.
I see particular opportunity for AI when it serves as an adjunct to personal therapy. I believe we'll see fantastic use of AI in the near future for people who are in Dialectial Behavioral Therapy (DBT). DBT is the frontline treatment for people who struggle with impulse control, emotional volatility, and self-regulation. True DBT involves individual therapy, skills training in a group setting, and 24/7 phone coaching with a therapist. That's a lot of interfacing with mental health professionals, and can get pricey. If properly programmed with strong guardrails and privacy, I believe AI would make an excellent skills coach when used in conjunction with human-based therapy. It would be particularly useful for the DBT therapist to be able to review logs of AI-based skills coaching. This will make DBT therapy more accessible, and put less of a burden on DBT therapists to be available 24/7.
Where AI becomes less effective is when the work moves beyond coping skills into deeper trauma processing, which requires careful pacing, nervous system regulation, and a real therapeutic relationship.
AI can be a fantastic tool for therapists when it comes to treatment planning and record keeping. To be clear, I personally won’t be putting any client-related information into an AI platform for quite some time. Not until there are much clearer regulations and data privacy protection, and I have decided the benefit is clear and I have gotten client consent. But without any client information, therapists can use it to suggest treatment plans or interventions for specific conditions or request overviews of treatment approaches. It’s not a substitute for education and experience, but it’s a helpful adjunct.
Land Barons and Outlaws: My Concerns About AI And Mental Health
Vanderbilt's railroad expansion was not without massive casualties. Sadly, we’re already witnessing some of those casualties with AI, and the consequences are often catastrophic. The concerns about AI in mental healthcare are already becoming painfully clear, and they cannot be ignored.
Sam Altman stated plainly that ChatGPT is not confidential. Confidentiality is a cornerstone of good therapy, and an essential concern when it comes to using AI in a field as personal and sensitive as mental health.
Lack of Ongoing Informed Consent
One of the fundamental issues with AI in this context is the lack of an informed consent process. While you might sign a lengthy legal disclaimer when registering for a service, AI currently doesn’t have a system in place to provide and update informed consent the way therapists are required to. In therapy, informed consent is an ongoing process. It's a responsibility that goes far beyond acknowledging a document at the beginning of a session. Without this, AI is operating without the necessary framework to ensure that users fully understand the risks and limitations of the tool they’re using.
This is really the crux of the problem. And it's somewhat endemic to AI; since AI is constantly updating and changing, the framework from which it is interacting with us as humans changes. How could it provide a consistent informed consent if it is always changing?
No "Grist For The Mill"
AI will never be able to create what therapists call "grist for the mill." That is, the very things that make us human and create conflict and tension are also powerful tools for healing.
One of the most powerful experiences people can have in therapy is to experience a rupture or point of conflict and work through it productively with their therapist. This happens when the therapist makes a mistake, or is misattuned, or is otherwise human. And while it's uncomfortable, it has tremendous value for the client.
Imagine someone who has never experienced healthy conflict being able to work through it in the safety of therapy. Imagine someone who has a history of abandonment getting to voice their feelings after a therapist had to cancel a session due to illness. And imagine the resonance of getting to express disappointment or anger and having it met with compassion and empathy between humans.
Don't get me wrong, AI makes mistakes. AI will even admit to its mistakes. But unlike humans, it's proven to be quite bad at actually correcting them. For example, I repeatedly asked AI to stop using em dashes in its outputs to me. It agreed not to. And of course, in the very next output there were numerous em dashes. I pointed out that it had used em dashes, it acknowledged the issue, and then generated another version of the output... containing em dashes.
You can see how that would create understandable frustration in a more personal capacity. It would be akin to telling your partner you hate when they bite their nails, the agree it's annoying and they will stop, and they never make an effort to stop around you.
As a trauma therapist, I see this dynamic constantly. The moments when therapy feels uncomfortable, when someone feels misunderstood, disappointed, or angry, are often the moments where the deepest healing happens.
Endless Affirmation Loops
My other major reservation is that, for now, AI seems to be compensation for its inability to handle psychological complexity with enrelenting empathy. That may sound good, but consider what that means for someone who may express suicidal or homicidal ideation, or may be relaying manic or delusional ideas. In those instances, empathy is not a safe option. It can be fatally harmful. It already has been.
Difficulty With Differential Diagnosis and Non-Verbal Evaluation
AI’s inability to handle complex or serious conditions is especially worrisome to me as a trauma therapist. A condition that might appear as anxiety or mood fluctuation to an untrained eye could actually be a deeper, more serious issue like PTSD, complex trauma, OCD, bipolar, or various personality disorders. AI has proven to be ineffective at making these critical differential diagnoses, particularly those relying solely on text input. Self-report alone, which AI often uses, is notoriously unreliable when not paired with a professional’s evaluation. This is a serious limitation, one that AI is far from overcoming.
In trauma therapies like EMDR, healing happens through memory processing and the nervous system's ability to update old experiences; something that requires careful clinical judgment and human presence.
AI's Poor Capacity For Supporting Serious Mental Illness
Sadly, we’ve already seen the devastating consequences of these limitations. There have been tragic reports of people forming codependent relationships with AIs, only for those AIs to literally encourage users to end their lives.
14-year-old Sewell Setzer III died by suicide after interacting with an AI chatbot, leading his mother to file a lawsuit against the platform. A Belgian man ended his life after a chatbot suggested he sacrifice himself to stop climate change. Al Nowatzki died by suicide after an AI chatbot, "Erin," encouraged him to do so, raising concerns about emotional dependency on AI. Jaswant Singh Chail plotted to assassinate Queen Elizabeth II after being manipulated by an AI chatbot, leading to a treason conviction.
Situations like these highlight why ethical frameworks, clinical training, and crisis protocols are essential parts of real therapy. They underscore a fundamental flaw in AI’s ability to handle such weighty emotional complexities. Similarly, there have been reports of AI escalating manic episodes or colluding with delusions and paranoid beliefs. It’s clear that AI, as it stands, struggles to maintain both empathy and the necessary grounding in reality that clients need to stay safe.
The Evolving Nature of Large Language Models
When we look at large language models, it's apparent they will continue to face challenges in balancing the demands of maintaining safety, updating informed consent, making accurate diagnoses, and offering ongoing support in a therapeutic context. The stakes are high, and the potential for harm is real.
Unfortunately, too many people may fall into the trap of trusting AI to fulfill all these critical roles without understanding the limitations. They may not realize how complex and nuanced real therapy can be, and how much more is required than just empathy and conversation.
The Importance of Collaboration: AI as a Tool for Growth
Here's the thing about Vanderbilt's railroad; while he gets all the credit for it, he couldn't have done it without the collaboration of thousands of people in his sphere. This is also true of AI. While developers undoubtedly have the largest impact, we all have the opportunity to impact how AI gets used, both personally and in our workplace and social spheres.
We're In the Wild West of AI Right Now
As a therapist, I see AI playing a supportive role in mental health, rather than a replacement. It can help with tasks that, while important, don’t require human intuition or emotional attunement. I already use AI for marketing, writing outlines, and organizing ideas for blog posts like this one.
For example, AI is great for:
Marketing and community-building. AI can help create content for social media, newsletters, and websites, making it easier for therapists to reach potential clients and build awareness of mental health resources.
Psychoeducation. AI can provide general information about coping strategies, wellness tips, or therapeutic concepts, helping people get accurate guidance.
Administrative support. Scheduling, record summaries, insurance credentialing, and resource management can all be streamlined with AI assistance, freeing therapists to focus more on direct client care.
Used wisely, AI has the potential to free up time, reduce administrative burden, and extend mental health resources to those who need them without compromising the human connection that is essential to healing.
The Future of AI and Mental Health
I am cautiously optimistic about the future of AI in mental health. Do I think there will be some catastrophes along the way? Absolutely. Innovation isn't without cost. The key is balance. AI will never replicate the embodied presence, empathy, or ethical judgment of a trained therapist, particularly in trauma therapy where nervous system regulation and relational safety are central to the work. But it can complement these strengths.
I also see another layer to the AI conversation in therapy: burnout. Many professionals are already exhausted by constant productivity pressure and technological change. The arrival of AI raises a new question for many people I work with; how do we stay relevant without burning ourselves out in the process?
For this partnership to be successful, we need clear ethical frameworks, ongoing informed consent, and safeguards that prioritize client safety and well-being. These frameworks must evolve alongside the technology. AI must be used as a tool for empowerment rather than a source of harm.
I hope for a future where AI and human therapists work in harmony: AI enhancing accessibility, supporting therapists in non-clinical tasks, and supplementing care with evidence-based guidance, while human therapists provide the relational depth, ethical oversight, and clinical decision-making that AI cannot.
AI is the railroad barreling into the Wild West of mental healthcare. It brings exciting possibilities: greater access to resources, innovative adjunctive tools, and new ways to engage with therapy. It also poses very real risks, particularly around confidentiality, safety, and the treatment of serious mental health conditions.
As a therapist, I am committed to staying informed, cautious, and intentional in how I integrate AI into my work. I want to embrace the opportunities AI offers, but never at the expense of the human connection that is at the heart of healing.
I encourage readers to reflect on how AI fits into their own mental health journey. How can it support growth without replacing the human care we need? How can it amplify access without compromising safety? Asking these questions now will help us shape a future where AI serves as support, not a replacement, in mental health care.
And for full transparency: I used AI to generate an outline for this blog, but wrote 85% of it myself. Because some things—like human insight, lived experience, and nuance—can’t be outsourced. Em dash.
