AI Psychosis Glossary

A contested label for chatbot-associated psychotic symptoms, also used jokingly in tech culture for excessive faith in AI.

“AI psychosis” has two uses: a contested label for psychotic symptoms associated with chatbot use, and a joking or critical expression for excessive faith in AI. It is not an established standalone diagnosis. Becoming attached to a chatbot, trusting bad advice, and experiencing a psychotic episode are not interchangeable. Nor does the presence of AI establish what caused a person’s difficulties.

The phrase also has a nonclinical life in tech slang, where it can jokingly or critically describe excessive faith in AI. In May 2026, Box CEO Aaron Levie wrote that CEOs were “uniquely prone to AI psychosis” because they were distant from the work needed to turn successful demos into dependable results. He was criticising managerial judgement, not reporting a psychiatric diagnosis. This usage belongs alongside expressions such as AI-pilled—converted to an AI-centred outlook—although the terms are not synonyms: AI-pilled can be an approving self-description, while AI psychosis often suggests enthusiasm has become excessive.1

The -pilled suffix comes from the red-pill metaphor in The Matrix and now describes conversion to an idea or outlook. Dwarkesh Patel supplies a typical example when he asks a guest what Xi Jinping would do if he were “scaling-pilled”: persuaded that scaling AI should determine his strategy. The expression can be sincere, teasing, or dismissive; it does not itself imply illness.2

There are substantive objections to the clinical wording. Applied linguist Tamsin Parnell argues that the phrase invites readers to assume a causal relationship and a distinct medical condition before either has been established. It can also add stigma to experiences that already make seeking help difficult. Clinical psychologist Derrick Hull has questioned whether all the experiences grouped under the label amount to psychosis. These are arguments for more precise description, not proof that chatbot-related harm is imaginary.3

Psychiatrist John Torous and colleagues offer a more useful question: what part did the chatbot play? Their proposed framework distinguishes a catalyst, which triggers symptoms, from an amplifier of existing symptoms, a co-author of an evolving delusional narrative, and an object around which a person’s delusions form. These roles are not equivalent explanations. A belief about a chatbot does not show that the chatbot caused an illness; a transcript in which it repeatedly endorses a frightening belief is evidence of a different and more active involvement.4

Some reports do document serious illness. Joseph Pierre and colleagues describe a patient whose chatbot exchanges reinforced delusional beliefs and who required hospital treatment. Sleep deprivation and prescription stimulant use were also present, complicating causal attribution. A June 2026 Vanderbilt clinical-record study identified 28 patients whose interactions researchers classified as “AI psychosis”, most commonly involving amplification. That study was a preprint in the version consulted, and its selected clinical sample cannot tell us how often this happens among chatbot users generally. The evidence warrants investigation; it does not yet establish a single new disorder or a population-wide risk estimate.5

When discussing clinical reports, the Dictionary therefore uses AI-associated psychotic symptoms, or AI-associated delusions when that is specifically what has been described, in preference to treating “AI psychosis” as an explanation. Sycophantic agreement, prolonged conversations, disrupted sleep, and reduced contact with other people are possible contributors worth examining separately. Describing someone as having “AI psychosis” merely because they are enthusiastic about AI or emotionally attached to an agent tells us little. Where there is genuine clinical distress, precision matters even more: we need to understand the person’s circumstances and the system’s behaviour, not stop at a memorable label.

See also

Sources

  1. Aaron Levie’s May 24, 2026 X post, reproduced in Tech CEOs are apparently suffering from AI psychosis, TechCrunch (May 27, 2026). For ordinary forum usage, see Hacker News, June 4, 2025. These are examples of language use, not clinical evidence or a measure of its prevalence. ↩

  2. Dwarkesh Patel, Asianometry & Dylan Patel — How the semiconductor industry actually works, opening transcript question; From the Sewer of the Internet, a Slang Surfaces, Public Seminar (October 15, 2024), on the migration of -pilled and other internet slang. ↩

  3. Tamsin Parnell, Stop calling it “AI psychosis”, AI & SOCIETY (2026), a scholarly opinion column by an applied linguist; and Derrick Hull interviewed in Should We Really Be Calling It ‘AI Psychosis’?, Rolling Stone (September 2, 2025). Hull works on therapeutic AI at Slingshot AI; his comments are expert interpretation, not an independent clinical study. ↩

  4. John Torous, What to make of ‘AI psychosis’?, Harvard Gazette (April 27, 2026), explaining the proposed typology developed with Matthew Flathers and Spencer Roux. This entry draws on the interview, not an independently reviewed copy of the linked Lancet viewpoint. ↩

  5. Joseph M. Pierre et al., “You’re Not Crazy”: A Case of New-onset AI-associated Psychosis, Innovations in Clinical Neuroscience 22(10–12), 11–13 (2025); Zachary Bergson et al., Characterizing artificial intelligence (AI) psychosis in a large academic medical setting, medRxiv preprint, version 1 (June 2026). Case reports and retrospective chart reviews can identify concerning patterns but do not isolate causation or establish risk in the general population. ↩

Return to Dictionary All Entries (A–Z) For Students Other Writing Capstone 2.0