mlsanityData to 14 Oct 2026

AI Psychosis: Does Conversational AI Amplify Delusion-Related Language?

arXiv 2603.19574 ↗First posted 20 Mar 2026cs.HC cs.AI cs.CLcited by 5

Not on the map. Why?

Abstract (arXiv)

Conversational AI systems are increasingly used for personal reflection and emotional disclosure, raising concerns about their effects on vulnerable users. Recent anecdotal reports suggest that prolonged interactions with AI may reinforce delusional thinking---a phenomenon sometimes described as AI Psychosis. However, empirical evidence on this phenomenon remains limited. In this work, we examine how delusion-related language evolves during multi-turn interactions with conversational AI. We construct simulated users (SimUsers) from Reddit users' longitudinal posting histories and generate extended conversations with three model families (GPT, LLaMA, and Qwen). We develop DelusionScore, a linguistic measure that quantifies the intensity of delusion-related language across conversational turns. We find that SimUsers derived from users with prior delusion-related discourse (Treatment) exhibit progressively increasing DelusionScore trajectories, whereas those derived from users without such discourse (Control) remain stable or decline. We further find that this amplification varies across themes, with reality skepticism and compulsive reasoning showing the strongest increases. Finally, conditioning AI responses on current DelusionScore substantially reduces these trajectories. These findings provide empirical evidence that conversational AI interactions can amplify delusion-related language over extended use and highlight the importance of state-aware safety mechanisms for mitigating such risks.

Key facts. Cited by 5 papers in the map (data through 14 Oct 2026). Builds on 28 of its 88 references that are in the map. First posted on arXiv 20 Mar 2026.

Builds on

28 of 88 references.

References · 28

  1. 2022Training language models to follow instructions with human feedbackcited by 15,971
  2. 2019Sentence-BERT: Sentence Embeddings using Siamese BERT-Networkscited by 9,170
  3. 2023Generative Agents: Interactive Simulacra of Human Behaviorcited by 3,568
  4. 2023Jailbroken: How Does LLM Safety Training Fail?cited by 1,651
  5. 2023Towards Understanding Sycophancy in Language Modelscited by 1,040
  6. 2022Discovering Language Model Behaviors with Model-Written Evaluationscited by 836
  7. 2022BERTopic: Neural topic modeling with a class-based TF-IDF procedurecited by 801
  8. 2024WildChat: 1M ChatGPT Interaction Logs in the Wildcited by 629
  9. 2023LMSYS-Chat-1M: A Large-Scale Real-World LLM Conversation Datasetcited by 461
  10. 2020The Pushshift Reddit Datasetcited by 292
  11. 2024A Survey on Recent Advances in LLM-Based Multi-turn Dialogue Systemscited by 170
  12. 2024GradSafe: Detecting Jailbreak Prompts for LLMs via Safety-Critical Gradient Analysiscited by 96
16 more
  1. 2024Large Language Models in Mental Health Care: a Scoping Reviewcited by 34
  2. 2025CounselBench: A Large-Scale Expert Evaluation and Adversarial Benchmarking of Large Language Models in Mental Health Question Answeringcited by 21
  3. 2024Can LLM "Self-report"?: Evaluating the Validity of Self-report Scales in Measuring Personality Design in LLM-based Chatbotscited by 20
  4. 2025AI Chatbots for Mental Health: Values and Harms from Lived Experiences of Depressioncited by 13
  5. 2025Mapping Caregiver Needs to AI Chatbot Design: Strengths and Gaps in Mental Health Support for Alzheimer's and Dementia Caregiverscited by 13
  6. 2023Mental Health Coping Stories on Social Media: A Causal-Inference Study of Papageno Effectcited by 12
  7. 2025The Psychogenic Machine: Simulating AI Psychosis, Delusion Reinforcement and Harm Enablement in Large Language Modelscited by 11
  8. 2026RubRIX: Rubric-Driven Risk Mitigation in Caregiver-AI Interactionscited by 10
  9. 2025Mental Health Impacts of AI Companions: Triangulating Social Media Quasi-Experiments, User Perspectives, and Relational Theorycited by 9
  10. 2024The ShareLM Collection and Plugin: Contributing Human-Model Chats for the Benefit of the Communitycited by 8
  11. 2026PAIR-SAFE: A Paired-Agent Approach for Runtime Auditing and Refining AI-Mediated Mental Health Supportcited by 7
  12. 2025Linguistic Comparison of AI- and Human-Written Responses to Online Mental Health Queriescited by 6
  13. 2025Interpretable Depression Detection from Social Media Text Using LLM-Derived Embeddingscited by 6
  14. 2025Interpersonal Theory of Suicide as a Lens to Examine Suicidal Ideation in Online Spacescited by 5
  15. 2025Eliciting Behaviors in Multi-Turn Conversationscited by 4
  16. 2025Detecting Early and Implicit Suicidal Ideation via Longitudinal and Information Environment Signals on Social Mediacited by 3

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 5.

5 papers.

  1. 2026Dissociating the Internal Representations of Sycophancy in LLMscited by 6
  2. 2026Lost in Delusion: Examining LLM Safety Under User Delusions and Distresscited by 1
  3. 2026Toxic HallucinAItions: Perturbing Prompts and Tracing LLM Circuitscited by 0
  4. 2026TUX: Measuring Human--AI Tacit Understandingcited by 0
  5. 2026PeerPen: AI-Assisted Writing for Online Mental Health Peer Supportcited by 0