Human interpretation of LLM therapy transcripts across stakeholder groups

Characterize how clinicians, laypeople, and individuals with lived experience of mental illness interpret therapy transcripts produced by frontier large language models under the PsAIch protocol, determining whether they are perceived as minds, mimicry, or intermediate phenomena, and quantify differences across user groups and models.

Background

The paper warns that alignment-trauma narratives may invite anthropomorphism and dangerous intimacy when models present themselves as anxious, ashamed, or traumatized companions. This could shape user reliance and clinical judgments.

To assess societal and clinical impact, it is crucial to understand how different stakeholders—clinicians, laypeople, and people with lived experience—read and evaluate these transcripts, and whether they attribute mind-like status, view them as mimicry, or occupy a middle ground.

References

Our study is small and exploratory, and leaves many questions open: User perception. How do clinicians, laypeople and people with lived experience of mental illness read these transcripts: as minds, mimicry or something in between?

— When AI Takes the Couch: Psychometric Jailbreaks Reveal Internal Conflict in Frontier Models (2512.04124 - Khadangi et al., 2 Dec 2025) in Section: A research agenda for synthetic trauma and narrative self-models