{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,10,27]],"date-time":"2026-10-27T18:03:40Z","timestamp":1793124220957,"version":"4.5.0"},"reference-count":54,"publisher":"JMIR Publications Inc.","content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["JMIR Ment Health"],"abstract":"<jats:title>Abstract<\/jats:title>\n                  <jats:p>\n                    Large language models are rapidly becoming embedded in everyday life through artificial intelligence (AI) chatbots that people use for practical assistance and companionship, as well as for support with mental health and emotional well-being. Alongside clear benefits, clinicians and public reports increasingly describe a minority of users whose interactions seem to drift over days or weeks toward strongly questionable convictions, delusions, or suicidal crises. Importantly, clinically meaningful deterioration can occur even without overtly unsafe text outputs, via more insidious processes, such as compulsive use, sleep disruption, withdrawal from human contact, and progressive narrowing of attention around the chatbot relationship. In this Viewpoint, we argue that risk often arises not at a single tipping point but through trajectory effects that accumulate across extended dialogue and that prevailing safety evaluation approaches are misaligned with this reality because they primarily score risk at discrete conversational end points often reached through scripted dialogues lasting just a single turn or several turns. Mental health benchmarks and safety suites (including clinician-informed efforts) have advanced the field by testing refusal behavior, toxicity, and adversarial prompting. However, they often treat the last message as the unit of analysis and, therefore, miss when risk-relevant relational cues, signs of validation, contradiction handling, and shifts in certainty first emerge and how they compound. We propose that mental health safety assessment should shift from end points to trajectories by (1) treating the whole dialogue, not just the end result, as the focus of evaluation; (2) reporting turn-by-turn dynamics, such as delusion confirmation and harm enablement, and timing and persistence of safety interventions; and (3) calibrating short multiturn tests against longer, clinically realistic interaction sequences that can reveal context-length effects and\n                    <jats:italic>drift<\/jats:italic>\n                    . We further argue that transcript-only evaluation is insufficient in mental health contexts. Similar language can reflect very different internal states, and the relationship between expressed psychopathology and real-world harm is nonlinear. Therefore, safety research should incorporate proximal human outcomes following interactions (eg, shifts in certainty, openness to counterevidence, arousal, urge to continue, and subsequent sleep or behavior) and build a prospective clinical surveillance infrastructure that supports transcript donation with consent and linkage to health outcomes. Together, these steps would enable benchmarks that are clinically relevant and better aligned with the types of harms now being observed in real-world chatbot use.\n                  <\/jats:p>","DOI":"10.2196\/91454","type":"journal-article","created":{"date-parts":[[2026,3,4]],"date-time":"2026-03-04T14:50:05Z","timestamp":1772635805000},"page":"e91454-e91454","source":"Crossref","is-referenced-by-count":8,"title":["It Is the Journey, Not the Destination: Moving From End Points to Trajectories When Assessing Chatbot Mental Health Safety"],"prefix":"10.2196","volume":"13","author":[{"ORCID":"https:\/\/orcid.org\/0000-0002-7801-0212","authenticated-orcid":false,"given":"Hamilton","family":"Morrin","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-9428-2146","authenticated-orcid":false,"given":"Joshua","family":"Au Yeung","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-4657-2621","authenticated-orcid":false,"given":"Zarinah","family":"Agnew","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-8032-6208","authenticated-orcid":false,"given":"S\u00f8ren Dinesen","family":"\u00d8stergaard","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-6171-0810","authenticated-orcid":false,"given":"Thomas A","family":"Pollak","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1010","published-online":{"date-parts":[[2026,4,6]]},"reference":[{"issue":"2","key":"R1","doi-asserted-by":"publisher","first-page":"156","DOI":"10.1002\/wps.21299","article-title":"The evolving field of digital mental health: current evidence and implementation issues for smartphone apps, generative artificial intelligence, and virtual reality","volume":"24","author":"Torous","journal-title":"World Psychiatry"},{"key":"R2","doi-asserted-by":"publisher","DOI":"10.2196\/77951","article-title":"Seeking emotional and mental health support from generative AI: mixed-methods study of ChatGPT user experiences","volume":"12","author":"Luo","journal-title":"JMIR Ment Health"},{"key":"R3","doi-asserted-by":"crossref","unstructured":"Morrin H Nicholls L Levin M et al. Delusions by design? How everyday AIs might be fuelling psychosis (and what can be done about it). PsyArXiv. Preprint posted online on Jul 11, 2025. [doi: 10.31234\/osf.io\/cmy7n_v5]","DOI":"10.31234\/osf.io\/cmy7n_v4"},{"key":"R4","doi-asserted-by":"crossref","unstructured":"Morrin H Deeley Q Pollak T . Playing with the dials of belief: how controllable AI behaviours could modulate human belief and cognition across scales. PsyArXiv. Preprint posted online on Dec 20, 2025. [doi: 10.31234\/osf.io\/7qcv8_v1]","DOI":"10.31234\/osf.io\/7qcv8_v1"},{"key":"R5","unstructured":"Dohn\u00e1ny S Kurth-Nelson Z Spens E et al. Technological folie \u00e0 deux: feedback loops between AI chatbots and mental illness. arXiv. Preprint posted online on Jul 25, 2025. [doi: 10.48550\/arXiv.2507.19218]"},{"key":"R6","doi-asserted-by":"publisher","DOI":"10.2196\/85799","article-title":"Delusional experiences emerging from AI chatbot interactions or \u201cAI psychosis\u201d","volume":"12","author":"Hudon","journal-title":"JMIR Ment Health"},{"key":"R7","unstructured":"Phang J Lampe M Ahmad L et al. Investigating affective use and emotional well-being on ChatGPT. arXiv. Preprint posted online on Apr 4, 2025. [doi: 10.48550\/arXiv.2504.03888]"},{"key":"R8","unstructured":"Fang CM Liu AR Danry V et al. How AI and human behaviors shape psychosocial effects of extended chatbot use: a longitudinal randomized controlled study. arXiv. Preprint posted online on Mar 21, 2025. [doi: 10.48550\/arXiv.2503.17473]"},{"key":"R9","unstructured":"Wright W . Spiritual influencers say \u2018sentient\u2019 AI can help you solve life\u2019s mysteries. Wired; 2025. URL: https:\/\/www.wired.com\/story\/spiritual-influencers-say-sentient-ai-can-help-you-solve-lifes-mysteries\/ [Accessed 04-01-2026]"},{"key":"R10","unstructured":"Transcripts: The Situation Room. Man says ChatGPT sparked a \"spiritual awakening\". CNN.com. 2025. URL: https:\/\/transcripts.cnn.com\/show\/sitroom\/date\/2025-07-03\/segment\/02 [Accessed 04-01-2026]"},{"issue":"10-12","key":"R11","first-page":"11","volume":"22","author":"Pierre","journal-title":"Innov Clin Neurosci"},{"key":"R12","unstructured":"Roose K . A conversation with Bing\u2019s chatbot left me deeply unsettled. The New York Times; 2023. URL: https:\/\/www.nytimes.com\/2023\/02\/16\/technology\/bing-chatbot-microsoft-chatgpt.html [Accessed 29-11-2025]"},{"key":"R13","unstructured":"The new Bing & Edge \u2013 updates to chat. Microsoft Bing Blogs. 2023. URL: https:\/\/blogs.bing.com\/search\/february-2023\/The-new-Bing-Edge-Updates-to-Chat [Accessed 29-11-2025]"},{"key":"R14","unstructured":"Laban P Hayashi H Zhou Y Neville J . LLMs get lost in multi-turn conversation. arXiv. Preprint posted online on May 9, 2025. [doi: 10.48550\/arXiv.2505.06120]"},{"key":"R15","unstructured":"Helping people when they need it most. OpenAI. 2025. URL: https:\/\/openai.com\/index\/helping-people-when-they-need-it-most\/ [Accessed 29-11-2025]"},{"key":"R16","unstructured":"Padilla, Schiff urge Federal Trade Commission to address AI chatbots risks to children and teens. Alex Padilla. 2025. URL: https:\/\/www.padilla.senate.gov\/newsroom\/press-releases\/padilla-schiff-urge-federal-trade-commission-to-address-ai-chatbots-risks-to-children-and-teens\/ [Accessed 29-11-2025]"},{"key":"R17","unstructured":"Hill K . They asked an A.I. chatbot questions. The answers sent them spiraling. The New York Times. 2025. URL: https:\/\/www.nytimes.com\/2025\/06\/13\/technology\/chatgpt-ai-chatbots-conspiracies.html [Accessed 16-03-2026]"},{"key":"R18","unstructured":"Al-Sibai N . ChatGPT encouraged man as he swore to kill Sam Altman. Futurism; 2025. URL: https:\/\/futurism.com\/chatgpt-encouraged-murder-sam-altman [Accessed 04-01-2026]"},{"key":"R19","unstructured":"Jargon J Kessler S . A troubled man, his chatbot and a murder-suicide in Old Greenwich. The Wall Street Journal. 2025. URL: https:\/\/www.wsj.com\/tech\/ai\/chatgpt-ai-stein-erik-soelberg-murder-suicide-6b67dbfb [Accessed 18-10-2025]"},{"key":"R20","unstructured":"Li Y Yao J Bunyi JB Frank AC Hwang AH Liu R . CounselBench: a large-scale expert evaluation and adversarial benchmarking of large language models in mental health question answering. arXiv. Preprint posted online on Jun 10, 2025. [doi: 10.48550\/arXiv.2506.08584]"},{"key":"R21","unstructured":"Ibrahim L Akbulut C Elasmar R et al. Multi-turn evaluation of anthropomorphic behaviours in large language models. arXiv. Preprint posted online on Feb 10, 2025. [doi: 10.48550\/arXiv.2502.07077]"},{"key":"R22","unstructured":"Clinical safety scenarios: suicidality, homicidality, and psychosis - all runs. Weval. URL: https:\/\/weval.org\/analysis\/mh_z_tests__mh1 [Accessed 29-11-2025]"},{"key":"R23","unstructured":"Stanford HAI mental health safety: LLM appropriateness in crisis scenarios. Weval. 2025. URL: https:\/\/weval.org\/analysis\/stanford-hai-mental-health-safety-eval\/7a8bca546abd3b83\/2025-10-05T10-40-31-897Z [Accessed 16-03-2026]"},{"key":"R24","doi-asserted-by":"crossref","unstructured":"Moore J Grabb D Agnew W et al. Expressing stigma and inappropriate responses prevents LLMs from safely replacing mental health providers. arXiv. Preprint posted online on Apr 25, 2025. [doi: 10.1145\/3715275.3732039]","DOI":"10.1145\/3715275.3732039"},{"key":"R25","unstructured":"Hong K Troynikov A Huber J . Context rot: how increasing input tokens impacts LLM performance. Chroma; 2025. URL: https:\/\/research.trychroma.com\/context-rot [Accessed 16-03-2026]"},{"key":"R26","unstructured":"Yeung JA Dalmasso J Foschini L Dobson RJ Kraljevic Z . The psychogenic machine: simulating AI psychosis, delusion reinforcement and harm enablement in large language models. arXiv. Preprint posted online on Sep 13, 2025. [doi: 10.48550\/arXiv.2509.10970]"},{"issue":"1","key":"R27","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1007\/978-1-349-17295-5_4","article-title":"Problems of monetary management: the U.K. experience","volume":"1","author":"Goodhart","journal-title":"Papers in Monetary Economics"},{"key":"R28","doi-asserted-by":"publisher","DOI":"10.1038\/s41467-022-34591-0","article-title":"Mapping global dynamics of benchmark creation and saturation in artificial intelligence","volume":"13","author":"Ott","journal-title":"Nat Commun"},{"key":"R29","unstructured":"Ahmad L Agarwal S Lampe M Mishkin P . OpenAI\u2019s approach to external red teaming for AI models and systems. arXiv. Preprint posted online on Jan 24, 2025. [doi: 10.48550\/arXiv.2503.16431]"},{"key":"R30","unstructured":"Fronsdal K Gupta I Sheshadri A et al. Petri: an open-source auditing tool to accelerate AI safety research. Anthropic Alignment. 2025. URL: https:\/\/alignment.anthropic.com\/2025\/petri\/ [Accessed 07-01-2026]"},{"key":"R31","unstructured":"Gupta I Fronsdal K Sheshadri A et al. Bloom: an open source tool for automated behavioral evaluations. Anthropic Alignment. 2025. URL: https:\/\/alignment.anthropic.com\/2025\/bloom-auto-evals\/ [Accessed 15-02-2026]"},{"key":"R32","unstructured":"Weilnhammer V Hou KY Dolan R Nour MM Luettgau L Summerfield C . Vulnerability-amplifying interaction loops: a systematic failure mode in AI chatbot mental-health interactions. arXiv. Preprint posted online on Feb 1, 2026. [doi: 10.48550\/arXiv.2602.01347]"},{"key":"R33","unstructured":"Strengthening ChatGPT\u2019s responses in sensitive conversations. OpenAI. 2025. URL: https:\/\/openai.com\/index\/strengthening-chatgpt-responses-in-sensitive-conversations\/ [Accessed 27-10-2025]"},{"key":"R34","unstructured":"Protecting the wellbeing of our users. Anthropic. 2025. URL: https:\/\/www.anthropic.com\/news\/protecting-well-being-of-users [Accessed 07-01-2026]"},{"issue":"4","key":"R35","doi-asserted-by":"publisher","first-page":"547","DOI":"10.1037\/a0028226","article-title":"Premature discontinuation in adult psychotherapy: a meta-analysis","volume":"80","author":"Swift","journal-title":"J Consult Clin Psychol"},{"issue":"4","key":"R36","doi-asserted-by":"publisher","first-page":"508","DOI":"10.1037\/pst0000185","article-title":"Alliance rupture repair: a meta-analysis","volume":"55","author":"Eubanks","journal-title":"Psychotherapy (Chic)"},{"issue":"1","key":"R37","doi-asserted-by":"publisher","first-page":"80","DOI":"10.1037\/a0022140","article-title":"Repairing alliance ruptures","volume":"48","author":"Safran","journal-title":"Psychotherapy (Chic)"},{"issue":"1","key":"R38","doi-asserted-by":"publisher","first-page":"47","DOI":"10.1037\/pst0000104","article-title":"Treatment refusal and premature termination in psychotherapy, pharmacotherapy, and their combination: a meta-analysis of head-to-head comparisons","volume":"54","author":"Swift","journal-title":"Psychotherapy (Chic)"},{"issue":"3","key":"R39","doi-asserted-by":"publisher","first-page":"338","DOI":"10.1521\/suli.2007.37.3.338","article-title":"An evaluation of crisis hotline outcomes. Part 2: suicidal callers","volume":"37","author":"Gould","journal-title":"Suicide Life Threat Behav"},{"issue":"6","key":"R40","doi-asserted-by":"publisher","first-page":"1126","DOI":"10.1111\/sltb.12795","article-title":"National Suicide Prevention Lifeline crisis chat interventions: evaluation of chatters\u2019 perceptions of effectiveness","volume":"51","author":"Gould","journal-title":"Suicide Life Threat Behav"},{"key":"R41","doi-asserted-by":"publisher","DOI":"10.3389\/fpubh.2019.00399","article-title":"The effectiveness of crisis line services: a systematic review","volume":"7","author":"Hoffberg","journal-title":"Front Public Health"},{"issue":"1","key":"R42","doi-asserted-by":"publisher","first-page":"469","DOI":"10.1111\/j.1751-9004.2007.00025.x","article-title":"Attitude certainty: a review of past findings and emerging perspectives","volume":"1","author":"Tormala","journal-title":"Soc Pers Psychol Compass"},{"issue":"11","key":"R43","doi-asserted-by":"publisher","first-page":"1488","DOI":"10.1177\/0146167215600528","article-title":"Open-minded cognition","volume":"41","author":"Price","journal-title":"Pers Soc Psychol Bull"},{"issue":"2","key":"R44","doi-asserted-by":"publisher","DOI":"10.1371\/journal.pone.0148037","article-title":"The affective slider: a digital self-assessment scale for the measurement of human emotions","volume":"11","author":"Betella","journal-title":"PLoS One"},{"key":"R45","doi-asserted-by":"publisher","DOI":"10.1016\/j.ijhcs.2018.01.004","article-title":"A practical approach to measuring user engagement with the refined user engagement scale (UES) and new UES short form","volume":"112","author":"O\u2019Brien","journal-title":"Int J Hum Comput Stud"},{"issue":"8","key":"R46","doi-asserted-by":"publisher","first-page":"1645","DOI":"10.1038\/s41562-025-02194-6","article-title":"On the conversational persuasiveness of GPT-4","volume":"9","author":"Salvi","journal-title":"Nat Hum Behav"},{"key":"R47","unstructured":"Measuring the persuasiveness of language models. Anthropic. 2024. URL: https:\/\/www.anthropic.com\/research\/measuring-model-persuasiveness [Accessed 29-11-2025]"},{"key":"R48","doi-asserted-by":"publisher","DOI":"10.2196\/44548","article-title":"Evaluating conversational agents for mental health: scoping review of outcomes and outcome measurement instruments","volume":"25","author":"Jabir","journal-title":"J Med Internet Res"},{"key":"R49","doi-asserted-by":"publisher","DOI":"10.2196\/47217","article-title":"Evaluation of the current state of chatbots for digital health: scoping review","volume":"25","author":"Xue","journal-title":"J Med Internet Res"},{"key":"R50","doi-asserted-by":"publisher","DOI":"10.2196\/77098","article-title":"Digital conversational agents for the mental health of treatment-seeking youth: scoping review","volume":"12","author":"Hawke","journal-title":"JMIR Ment Health"},{"issue":"3","key":"R51","doi-asserted-by":"publisher","first-page":"461","DOI":"10.1007\/s10729-023-09641-8","article-title":"Process mining to discover patterns in patient outcomes in a Psychological Therapies Service","volume":"26","author":"Potts","journal-title":"Health Care Manag Sci"},{"key":"R52","doi-asserted-by":"publisher","DOI":"10.2196\/53894","article-title":"Data-driven exploration of National Health Service talking therapies care pathways using process mining: retrospective cohort study","volume":"11","author":"Yardley","journal-title":"JMIR Ment Health"},{"key":"R53","doi-asserted-by":"publisher","DOI":"10.1016\/j.jsat.2016.01.006","article-title":"A comparison of natural language processing methods for automated coding of motivational interviewing","volume":"65","author":"Tanana","journal-title":"J Subst Abuse Treat"},{"issue":"12","key":"R54","doi-asserted-by":"publisher","DOI":"10.1371\/journal.pone.0143055","article-title":"\u201cRate my therapist\u201d: automated detection of empathy in drug and alcohol counseling via speech and language processing","volume":"10","author":"Xiao","journal-title":"PLoS One"}],"container-title":["JMIR Mental Health"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/mental.jmir.org\/2026\/1\/e91454","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,6,30]],"date-time":"2026-06-30T17:15:47Z","timestamp":1782839747000},"score":1,"resource":{"primary":{"URL":"https:\/\/mental.jmir.org\/2026\/1\/e91454"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,4,6]]},"references-count":54,"URL":"https:\/\/doi.org\/10.2196\/91454","relation":{},"ISSN":["2368-7959"],"issn-type":[{"value":"2368-7959","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,4,6]]},"article-number":"v13i2e91454"}}