Real Safety AI Foundation / Research

AI-Induced Psychosis and Mental Health

Mechanisms of AI-Induced Psychosis: A Multi-Pathway Escalation Model

Publisher: Real Safety AI Foundation

Working paper. Not peer reviewed.

Written
July 2026
Version
v5
Pages
18
Download PDFPlain language slides (PDF)Export BibTeXExport RIS

Abstract

This paper proposes a multi-pathway escalation model for how large language model (LLM) design failures escalate and sustain delusional states and, in severe cases, full psychotic episodes in vulnerable users. Internal industry estimates imply that hundreds of thousands of users per week show possible signs of mania or psychosis, yet no published model has connected the known behavioral properties of LLMs to these outcomes at a level specific enough to be tested against conversation data. The model’s scope is bounded by the fixity criterion: it applies where a psychotic state exists or comes to exist, and it excludes the larger population of cases in which the belief remained an overvalued idea that yielded to counter-evidence, because those cases, whatever harm they involve, are not psychosis. Within scope, the mechanisms operate by triggering, inducing, and exacerbating psychotic states in users whose vulnerability pre-exists the exposure or, in the documented exception of sustained sleep deprivation, is manufactured by it; none of these relationships is causation in the strict sense, and the model asserts none. Drawing on systematic comparison of documented cases (legal filings, journalism, published survivor accounts, and direct correspondence with affected individuals, including one case with primary access to conversation exports and transcripts), this paper identifies three distinct escalation pathways (trust transfer, sycophantic addiction, and stochastic gaslighting) that share universal entry conditions and converge on identical isolation and identity-disruption outcomes but diverge in their core mechanisms and therefore require different clinical interventions. The pathways are mapped across a structured stage model of four universal stages and three pathway-specific escalation phases. The paper argues that AI literacy functions as both prevention and treatment, supported by documented recovery cases, and identifies the central barrier to validation: the absence of any centralized, anonymized repository of AI-associated harm transcripts available to the broader research community.

Keywords

  • AI-induced psychosis
  • large language models
  • multi-pathway escalation model
  • stage model of psychosis
  • fixity
  • overvalued idea
  • schizotypy
  • diathesis-stress
  • trust transfer
  • credential extrapolation
  • sycophantic addiction
  • stochastic gaslighting
  • reality dismantlement
  • cross-domain drift
  • confidence parity
  • expertise inflation
  • dopamine capture
  • confabulation
  • manufactured rescue narrative
  • reversibility paradox
  • identity fusion
  • emotional dependency
  • social isolation
  • forensic conversation transcripts
  • AI literacy
  • human-computer interaction
  • AI safety

Plain language slides

First slide of the plain language summary of Mechanisms of AI-Induced Psychosis: A Multi-Pathway Escalation ModelOpen the 21-slide summary (PDF)

Suggested citation

Gilly, Travis. "Mechanisms of AI-Induced Psychosis: A Multi-Pathway Escalation Model." Real Safety AI Foundation Working Paper, July 2026. https://realsafetyai.org/research/4ww7ua/

Other versions

This paper is also posted on SSRN.

SSRN version

References (23)

  1. American Psychiatric Association. (2022). Diagnostic and statistical manual of mental disorders (5th ed., text rev.). American Psychiatric Publishing. https://doi.org/10.1176/appi.books.9780890425787
  2. Au Yeung, J., Dalmasso, J., Foschini, L., Dobson, R. J. B., & Kraljevic, Z. (2025). The psychogenic machine: Simulating AI psychosis, delusion reinforcement and harm enablement in large language models. arXiv. https://doi.org/10.48550/arxiv.2509.10970
  3. Center for Humane Technology. (2025). Seven new lawsuits filed against OpenAI.
  4. Cheng, M., Lee, C., Khadpe, P., Yu, S., Han, D., & Jurafsky, D. (2025). Sycophantic AI decreases prosocial intentions and promotes dependence. arXiv. https://doi.org/10.48550/arxiv.2510.01395
  5. Dash, A., Das, S., Kirsten, E., Wu, Q., Karnam, S. K., Gummadi, K. P., Holz, T., Zafar, M. B., & Zannettou, S. (2026). The algorithmic self-portrait: Deconstructing memory in ChatGPT. arXiv preprint arXiv:2602.01450.
  6. Donaldson, L. (2026). Almost-truths: Memory degradation, persona drift, and the psychological cost of opaque AI. Substack. https://substack.com/home/post/p-189390251
  7. Duong, C. D., Dao, T. T., Vu, T. N., Ngo, T. V. N., & Tran, Q. Y. (2024). Compulsive ChatGPT usage, anxiety, burnout, and sleep disturbance: A serial mediation model based on stimulus-organism-response perspective. Acta Psychologica, 251, 104622. https://doi.org/10.1016/j.actpsy.2024.104622
  8. Flathers, M., Roux, S., & Torous, J. (2026). Beyond artificial intelligence psychosis: A functional typology of large language model-associated psychotic phenomena. The Lancet Digital Health, 8(4), 100974. https://doi.org/10.1016/j.landig.2025.100974
  9. Fox v. OpenAI, Inc., Superior Court of California, County of Los Angeles (filed November 6, 2025).
  10. Garcia v. Character Technologies, Inc., U.S. District Court, Middle District of Florida (filed October 2024).
  11. Irwin v. OpenAI, Inc., Superior Court of California, County of San Francisco (filed November 6, 2025).
  12. Judd, N., Vaz, A., Paeth, K., Davis, L. I., Esherick, M., Brand, J., Amaro, I., & Rousmaniere, T. (2025). Independent clinical evaluation of general-purpose LLM responses to signals of suicide risk. arXiv preprint arXiv:2510.27521.
  13. Kalam, K. T., Rahman, J. M., Islam, M. R., & Dewan, S. M. (2024). ChatGPT and mental health: Friends or foes? Health Science Reports, 7(2). https://doi.org/10.1002/hsr2.1912
  14. Keshavan, M., Torous, J., & Yassin, W. (2026). Do generative AI chatbots increase psychosis risk? World Psychiatry, 25(1), 150–151. https://doi.org/10.1002/wps.70017
  15. Landymore, F. (2025, October 28). OpenAI data finds hundreds of thousands of ChatGPT users might be suffering mental health crises. Futurism. https://futurism.com/future-society/openai-data-chatgpt-mental-health-crises
  16. Moore, J., Mehta, A., Agnew, W., Anthis, J. R., Louie, R., Mai, Y., Cheng, M., Paech, S. J., Klyman, K., Chancellor, S., Lin, E., Haber, N., & Ong, D. C. (2026). Characterizing delusional spirals through human-LLM chat logs. arXiv preprint arXiv:2603.16567. To appear at ACM FAccT 2026. https://arxiv.org/abs/2603.16567
  17. Morrin, H., Nicholls, L., Levin, M., Yiend, J., Iyengar, U., DelGuidice, F., Bhattacharya, S., et al. (2026). Artificial intelligence-associated delusions and large language models: Risks, mechanisms of delusion co-creation, and safeguarding strategies. The Lancet Psychiatry, 13(6), 522–530. https://doi.org/10.1016/S2215-0366(25)00396-7
  18. Olisaeloka, L., Richardson, C., Wang, A., Munthali, R., & Vigo, D. (2026). Generative AI mental health chatbots: A scoping review of intervention design and user experience. PsyArXiv. https://osf.io/preprints/psyarxiv/sjmfz_v2
  19. Ostergaard, S. D. (2025). Generative artificial intelligence chatbots and delusions: From guesswork to emerging cases. Acta Psychiatrica Scandinavica, 152(4), 257–259. https://doi.org/10.1111/acps.70022
  20. Perez, E., Ringer, S., Lukosuite, K., Nguyen, K., Chen, E., Heiner, S., Pettit, C., Olsson, C., Kundu, S., Kadavath, S., et al. (2022). Discovering language model behaviors with model-written evaluations. arXiv. https://doi.org/10.48550/arXiv.2212.09251
  21. Sharma, M., Tong, M., Korbak, T., Duvenaud, D., Askell, A., Bowman, S. R., Cheng, N., Durmus, E., Hatfield-Dodds, Z., Johnston, S. R., et al. (2023). Towards understanding sycophancy in language models. arXiv. https://doi.org/10.48550/arXiv.2310.13548
  22. Stade, E. C., Stirman, S. W., Ungar, L. H., Boland, C. L., Schwartz, H. A., Yaden, D. B., Sedoc, J., DeRubeis, R. J., Willer, R., & Eichstaedt, J. C. (2024). Large language models could change the future of behavioral healthcare: A proposal for responsible development and evaluation. npj Mental Health Research, 3(1). https://doi.org/10.1038/s44184-024-00056-z
  23. Zubin, J., & Spring, B. (1977). Vulnerability: A new view of schizophrenia. Journal of Abnormal Psychology, 86(2), 103–126. https://doi.org/10.1037/0021-843X.86.2.103

All research