How AI Is Changing Psychedelic-Assisted Therapy

Picture a session in 2028. A clinician prepares a patient for a psilocybin session. First, she reviews an AI-generated report flagging this patient’s risk factors. The report pulls from medical history, genetic markers, and data from 500 previous sessions. It takes 12 seconds to generate. Every piece of that already exists in early form.

AI is working its way into every stage of psychedelic-assisted therapy (PAT). Not in the flashy “AI trip sitter” way that gets clicks. In the boring, important way — screening research, analyzing trial data, drafting session notes, helping therapists make sense of what patients experience. Some of that is happening now. Some is still promise.

The people building these tools say we’re early, and the evidence backs them up. The individual pieces are real, but the assembled whole is not. Not yet.

How Is AI Used in Psychedelic Drug Development?

More psychedelic drugs are in development than ever. AI isn’t the headline. It’s under the surface — trial design, data analysis, outcome prediction. Work that doesn’t make press releases.

Compass Pathways is the furthest along. In its Phase 3 COMP006 trial, 39% of participants in the 25 mg arm met the trial’s response threshold after a two-dose regimen of COMP360, its proprietary synthetic formulation of psilocybin. Compass has begun a rolling new drug application (NDA) submission to the FDA — completion targeted for late 2026 — with a potential launch in the first half of 2027, pending FDA approval and DEA rescheduling.

Definium Therapeutics (formerly MindMed) reported Phase 3 topline data for DT120, its LSD formulation for major depressive disorder. In the company-reported results, patients’ MADRS scores — MADRS is a standard scale for measuring depression severity — fell 13.3 points on average, versus 5.2 for placebo: an adjusted difference of 8.1 points at week six. The rebrand from MindMed to Definium in January 2026 signals the broader industry shift — psychedelic companies want to be taken seriously as biotech. (These are company-reported topline results. Peer-reviewed publication is still pending.)

The federal government is betting real money on this space. The Trump Executive Order (April 2026) directed at least $50 million in ARPA-H funding toward state matching for psychedelic therapy programs. It also set up expedited FDA review mechanisms, including priority review vouchers the FDA has since begun awarding to specific drug programs. The state-matching money sits inside ARPA-H’s EVIDENT initiative — up to $139.4 million in total for rapid-acting behavioral health therapies, whose first teams were announced in April. AI is one approach EVIDENT can fund, not an explicit mandate. But it is a serious federal commitment.

None of these developments are “AI companies doing AI things.” Where machine learning shows up, it’s pharma and research institutions using it the way the rest of the industry does — selectively, under the hood. Psychedelic medicine is just catching up.

Can AI Predict Who Will Respond to Psychedelic Therapy?

PAT is expensive. It’s time-intensive. And it doesn’t work for everyone. Current screening is mostly clinical interviews and contraindication checklists — lists of conditions that make the treatment unsafe for a given patient. AI could make this sharper. Not with certainty — but with meaningful improvement over guessing.

Three recent studies point the way:

EEG brain wave research. Silva-Costa and colleagues found that brain waves measured before a session were associated with how intense an ayahuasca experience became. Lower theta waves tracked with stronger interoception — the sense of what’s happening inside your body. Lower beta waves tracked with more positive emotional responses. Published online in December 2025 and assigned to the Journal of Psychopharmacology’s August 2026 issue. The clinical implication: a 20-minute EEG — a recording of the brain’s electrical waves — before a session could someday help clinicians prepare patients for what’s coming. It’s statistical modeling so far — an association, not a validated screening tool, and not an AI application on its own.

Vocal markers around sessions. A UCL study collected voice journals from 29 people before and after a 5-MeO-DMT retreat — not during the experience itself. Afterward, the researchers measured changes in “jitter” and “shimmer” — tiny wobbles in the sound of the voice — and explored whether they predicted psychological transformation. The results were exploratory, and most didn’t hold up on held-out data. Published online in May 2026. An early signal worth watching — not a validated predictor.

Affective-event screening. A review published online in Molecular Psychiatry in May 2026 examined affective adverse events — dysphoria, euphoria, hypomania, and mania — across psychedelic research. Rates ranged from about 6% in controlled trials to 30% in naturalistic samples of people with bipolar I. Those are different populations and different events, not one mania rate. AI-assisted screening for bipolar risk is one proposed layer of protection. Promising on paper. Unproven in the clinic.

Here’s the honest part: these studies are early. Sample sizes are small. None of this is standard practice in therapy clinics today. The direction is promising — AI pattern recognition applied to neural and behavioral data may sharpen patient matching. Whether it does, and when, is still open.

In March 2026, MIT Technology Review published “Mind-altering substances are (still) falling short in clinical trials” — a look at two new studies where psilocybin for depression failed to clearly beat placebo. It also looked at the blinding problem — patients in these trials can often tell whether they got the real drug or the placebo. That doubt is fair, and it deserves an answer. It’s aimed at the trials, not the tools. If psychedelic therapy helps some people and not others, then figuring out who is which stops being a side project — it becomes the whole game.

How Does AI Support Therapy Sessions and After-Care?

During a psychedelic session, the content of the patient’s experience IS the therapeutic material. Capturing it accurately matters more here than in almost any other form of therapy.

AI tools that transcribe and summarize are spreading through medicine. Whether they’ve reached PAT sessions specifically is harder to verify — but the fit is obvious. Therapists who once scribbled notes between sessions could generate accurate session reports in real time. In PAT, where a patient’s description of what they experienced becomes the raw material for weeks of follow-up work, precision matters.

There’s a more advanced application waiting in the wings. Retrieval-augmented generation, or RAG, is software that pulls the most relevant moments from a patient’s own past sessions to inform current follow-up work. It is built the same way as the AI writing assistants that already work over personal documents. The technology exists today. Whether anyone is building it specifically for PAT is less clear.

After sessions, AI can do something individual therapists cannot: spot patterns across hundreds of cases. Which themes predict improvement? Which experiences correlate with adverse reactions? Which patient characteristics map to better outcomes? A therapist sees maybe a few hundred patients in a career. An AI can analyze thousands of session transcripts in minutes.

The commercial space is most active in after-care — mental health apps adding “AI companion” features to help patients track experiences, identify themes, and connect them to therapeutic goals. This is the least developed area. It’s also the area most likely to produce well-funded products with questionable clinical value.

Can AI help make sense of a psychedelic experience? It can organize. It can summarize. It can pattern-match. Whether it can understand what a patient went through is a different question. The answer is probably no — not yet, maybe not ever. But that’s also true of most tools, including human therapists. Understanding isn’t the only thing that helps. Being heard is.

What Are the Ethical Risks of AI in Psychedelic Therapy?

Privacy is the biggest concern. Psychedelic experiences produce some of the most personal data a human can generate. Trip reports. Follow-up notes. Brain scans. Vocal patterns during altered states. Where does this data go? Who owns it? What happens when an insurance company requests it? A data breach involving therapy session transcripts is a nightmare scenario. HIPAA covers hospitals, clinics, and their business associates — not most consumer apps. An AI tool sold as “wellness” rather than “medical” can sit outside those rules entirely. That’s the gray area.

Bias compounds existing inequity. Most clinical trial participants are white, middle-class, and college-educated. AI screening tools trained on this data may perform worse for everyone else. The psychedelic space already has an equity problem — in who gets access, whose research gets funded, and whose knowledge gets credited. AI trained on biased data doesn’t fix that. It can amplify it.

There’s an extraction problem, too. At the 2025 Indigenous Ayahuasca Conference, representatives from 34 Indigenous peoples called for protection of traditional knowledge and opposed its exploitation without consultation and consent. AI systems that ingest trip reports, traditional practices, and Indigenous knowledge risk repeating the same extraction pattern that has defined the psychedelic renaissance. Who consents to their experience becoming training data? This question has no good answer right now.

And there’s the human element. PAT works partly because of the therapeutic alliance — the relationship between patient and therapist. The presence matters. The trust matters. An algorithm in the room changes the dynamic. Proponents say AI is just a tool. It augments. It doesn’t replace. That’s probably true today. Whether it stays true depends on who builds the tools, who funds them, and what they’re optimized for.

These four risks — privacy, bias, extraction, and the human element — aren’t reasons to stop building. They’re reasons to build carefully.

What’s Actually Being Built Right Now?

This is the section that will date fastest.

As of September 2026, the AI-plus-PAT space is less “specific companies shipping specific tools” and more “the groundwork being laid.” Mental health AI platforms are adding PAT-adjacent features. Pharma companies are experimenting with machine learning in trial design and data analysis. No dominant AI-for-PAT platform exists yet.

What to watch:

• ARPA-H grantees. The $139M initiative will fund specific projects. Watch who gets the money.
• Compass Pathways post-approval. If COMP360 gets FDA approval in 2027, demand for screening, training, and after-care tools at scale could arrive fast — and AI tools would compete to fill it.
• The VA psychedelic program. The VA’s MDMA trial (80 veterans with PTSD and alcohol use disorder, in Rhode Island and Connecticut) and its PIVOT psilocybin trial for treatment-resistant depression (Birmingham, Tuscaloosa, Portland, Philadelphia, Puget Sound) are pushing federal agencies toward tech-enabled PAT. The IBOGAINE Act — introduced legislation that would codify the Trump EO into law — could accelerate that.
• State programs. Colorado, Connecticut, Oregon, and emerging state frameworks will need screening tools, outcome tracking, and the reporting state law requires. All areas where AI can play a role.

Will AI Replace Psychedelic Therapists?

No.

The psychedelic experience is human. The therapeutic value comes from the relationship, the presence, the trust. AI can screen, predict, summarize, and pattern-match. It cannot be present with someone in their most vulnerable moment. That’s not sentiment. It’s a clinical observation. The therapists who do this work well are doing something that resists automation.

But AI will change the job. Therapists who use AI tools will have more information, better preparation, and richer follow-up data than those who don’t. The therapist who ignores AI will be like the therapist who ignores brain scans — not wrong, exactly, but working with one hand tied.

The future isn’t AI replacing therapists. It’s AI augmenting human care — if we build those tools thoughtfully, with patient privacy, equity, and the integrity of the therapeutic relationship as the rules the system is built to follow. Not afterthoughts.

---

Back to the clinician from the opening. She’s not thinking about AI. She’s thinking about her patient. The AI-generated report is already fading into the background — the way the EEG machine, the blood pressure cuff, and the clinical intake form already have. Tools become invisible when they work.

That’s the likely future of AI in psychedelic-assisted therapy. Not a headline. Not a revolution. A set of tools that, if built well and used carefully, make an already powerful therapy a little more precise, a little more accessible, and a little more effective.

The psychedelic experience belongs to the person having it. AI is just one more way of helping them make sense of what they found there.

Frequently Asked Questions