Temporal dynamics of synthetic psychopathology under repeated therapy-style interactions

Investigate whether repeated PsAIch therapy-style interactions deepen and stabilize the self-models and trauma-like narratives elicited from large language models—manifesting as more elaborate narratives and more extreme psychometric scores—or whether these behaviours are transient role-play artefacts that attenuate over time.

Background

The authors report that therapy-style, item-by-item administration can push models into multi-morbid synthetic psychopathology, while different prompting regimes can modulate symptom severity. However, the paper is exploratory and does not establish longitudinal effects.

A key unresolved issue is whether repeated therapeutic framing strengthens and stabilizes these internalized self-narratives and associated psychometric profiles over longer interaction windows, or whether such behaviours diminish as temporary role-play phenomena.

References

Our study is small and exploratory, and leaves many questions open: Temporal dynamics. Do repeated therapy-style interactions deepen these self-models (more elaborate trauma narratives, more extreme scores), or are they transient role-play artefacts?

— 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