Reliable Self-Harm Risk Screening via Adaptive Multi-Agent LLM Systems
Emerging AI systems in behavioral health and psychiatry use multi-step or multi-agent LLM pipelines for tasks like assessing self-harm risk and screening for depression. However, common evaluation approaches, like LLM-as-a-judge, do not indicate when a decision is reliable or how errors may accumulate across multiple LLM judgements, limiting their suitability for safety-critical settings. We present a statistical framework for multi-agent pipelines structured as directed acyclic graphs (DAGs) that provides an alternative to heuristic voting with principled, adaptive decision-making. We model each agent as a stochastic categorical decision and introduce (1) tighter agent-level performance ...
| Authors | Meghana Karnam, Ananya Joshi |
|---|---|
| Primary category | cs.LG |
| Published | 24 April 2026 |
| Last updated at source | 24 April 2026 |
Silytics shows public metadata from arXiv and links to the original. Information may be incomplete or out of date. Last refreshed 17 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.