Characterizing Delusional Spirals through Human-LLM Chat Logs
arXiv (preprint), 2026 · preprint, not peer reviewed
L5of 5
- Design
- Case series / log analysis
- Risk of bias
- High risk of bias
- Sample
- n = 19
- AI type
- General-purpose assistants
- Outcomes
- Psychosis
Authors' conclusion
“We conclude with concrete recommendations for how policymakers, LLM chatbot developers, and users can use our inventory and conversation analysis tool to understand and mitigate harm from LLM chatbots. Warning: This paper discusses self-harm, trauma, and violence.”Quoted verbatim from the abstract.
Draft grade. Level and risk of bias were assigned by our agent and await expert review. How we grade
Read the paper →https://arxiv.org/abs/2603.16567v1