When the therapist hallucinates: AI, psychedelics, and the risks of unsupervised digital mental health care.

As direct-to-consumer ketamine prescriptions and psychedelic support expand via telehealth, some patients are using general-purpose chatbots as de facto therapists during unmonitored altered-state sessions. This convergence of psychedelic therapeutics and generative AI creates an unregulated frontier in mental health care. Current AI tools are not licensed professionals, lack consistent reliability, cannot provide emergency support, and their unpredictability combined with mind-altering drugs can precipitate harm. Traditional safeguards relying on trained humans for supportive mindset and environment are disrupted when AI replaces that role. Clinicians should recognize patients already experiment with AI in this way, often without disclosure, and should counsel about dangers, document safety discussions, and routinely ask about AI use as part of risk assessment.

Study at a glance

Abstract

As direct-to-consumer ketamine prescriptions obtained via telehealth in the United States, along with psychedelic support, proliferate, some patients are turning to general-purpose chatbots as de facto "therapists" in unmonitored settings. These AI-guided altered-state sessions carry significant ethical, clinical, and legal risks. Current AI tools are not licensed professionals, lack consistent reliability, and cannot provide emergency support. Combined with mind-altering drugs, their unpredictability can precipitate harm rather than relieve distress. The convergence of these innovations (psychedelic therapeutics and generative AI) creates an unregulated frontier in mental health care. Traditional safeguards in psychedelic therapy rely on "set and setting," a supportive mindset and environment guided by trained humans. When that role is replaced by AI, both are fundamentally disrupted. Clinicians should recognize that patients are already experimenting with AI in this way, often without disclosure or oversight. When screening for psychedelic use or discussing at-home treatment, providers should anticipate the risks of inadequate monitoring, unsafe crisis response, and blurred therapeutic boundaries with AI chatbots. Even if clinicians do not endorse or recommend AI tools, they may bear responsibility once patient use is known. Providers should therefore counsel patients about these dangers, document safety discussions, and routinely ask about AI use as part of risk assessment.