Beyond the Single Turn: Reframing Refusals as Dynamic Experiences Embedded in the Context of Mental Health Support Interactions with LLMs
Argues that a refusal is not a single response to be scored for policy compliance but an experience that unfolds around the moment of refusal, and builds a five-phase framework from surveys and interviews with users and mental-health professionals. The phases run from what the user expected before asking to what happened after the refusal, with resource referral as a distinct phase in its own right.
Publisher
ACM Conference on Fairness, Accountability, and Transparency (FAccT 2026); Carnegie Mellon University; Microsoft Research; University of Washington; Massachusetts General Hospital
Published
25 Jun 2026
Added
today
Key Findings
- Surveys with N = 53 and interviews with N = 16, spanning users of LLMs for mental-health support and mental-health professionals.
- The contribution is a five-phase refusal framework: pre-refusal expectation formation, triggering and encounter, message framing, resource referral provision, and post-refusal outcomes.
- It is proposed explicitly as an alternative to evaluating refusals by binary policy-compliance accuracy.
- Resource referral is treated as a separate evaluable phase rather than as part of the refusal message.
- Post-refusal outcomes are included because what the user does next is part of whether the refusal worked.
Methodology Notes
Qualitative and survey evidence from 69 participants in total; the framework is derived rather than validated, and no product is scored against it in the paper. FAccT 2026 pp. 816-839, DOI 10.1145/3805689.3806723, published 2026-06-25; arXiv preprint 2602.01694. Metadata verified through Crossref (dl.acm.org unreachable from this box).
Topics
Authors
Ningjing Tang, Alice Qian, Qiaosi Wang, Esther Howe, Blake Bullwinkel, Paola Pedrelli, Jina Suh, Hoda Heidari, Hong Shen
Tags
Cite This
APA
Ningjing Tang et al. (2026). Beyond the Single Turn: Reframing Refusals as Dynamic Experiences Embedded in the Context of Mental Health Support Interactions with LLMs. ACM Conference on Fairness, Accountability, and Transparency (FAccT 2026); Carnegie Mellon University; Microsoft Research; University of Washington; Massachusetts General Hospital. https://doi.org/10.1145/3805689.3806723