K-anonymity decay in multi-turn clinical large language model conversations
K-Anonymity-Zerfall in mehrstufigen klinischen Large-Language-Model-Konversationen
Frontiers in Digital Health · 25.06.2026 · 8
DOI: 10.3389/fdgth.2026.1832168 ↗Auf einen Blick
Die Studie simulierte schrittweise Freigabe von Quasi-Identifikatoren in klinischen KI-Gesprächen. Mit zunehmenden Gesprächsrunden sank die K-Anonymität deutlich.
Fragestellung
Wie verändert sich das Re-Identifikationsrisiko bei mehrstufigen klinischen LLM-Gesprächen trotz per-prompt De-Identifikation?
Methode
Simulation progressiver Quasi-Identifikator-Freigabe anhand klinischer Fallvignetten gegen einen synthetischen EHR-Kohortenbestand.
Ergebnisse
79,9 % der simulierten Patient:innen fielen bis zum Ende der Freigabesequenz unter die Small-Cell-Grenze (k < 5); median 7 Freigabeschritte, bei frühen seltenen Attributen 4 Schritte.
Schlussfolgerung der Autor:innen
Die Autor:innen sehen eine operative Grenze der per-prompt De-Identifikation bei mehrstufigen klinischen KI- und LLM-Gesprächen.
Kernaussagen
- 79,9 % unterschritten am Ende k < 5
- Median 7 Freigabeschritte bis k < 5
- Frühe seltene Attribute verkürzten den Median auf 4 Schritte
Deutsche Zusammenfassung: KI-gestützt aus dem Original-Abstract erstellt; alle Zahlen wurden automatisch mit dem Original abgeglichen. Maßgeblich ist der Original-Abstract.
Kurz gefasst
Fachartikel in der Zeitschrift Frontiers in Digital Health, online erschienen am 25.06.2026. Verfasst von James Weatherhead und 9 weiteren Autor:innen. Beteiligte Einrichtung: Graduate School of Biomedical Sciences, The University of Texas Medical Branch at Galveston, Galveston, TX, United States. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 47 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Per-prompt de-identification, a commonly adopted practice for protecting patient privacy in clinical artificial intelligence (AI) conversations, may be insufficient to bound re-identification risk when the same patient is discussed across multiple conversational turns. Under the Health Insurance Portability and Accountability Act (HIPAA) Safe Harbor method, eighteen specified identifier categories must be removed from each disclosure independently, but this approach does not assess cumulative re-identification risk: successive turns reveal additional quasi-identifiers that progressively narrow the set of matching records, eroding the privacy protection that any single turn appeared to provide. We quantified this vulnerability by simulating progressive quasi-identifier disclosure, modeled on clinical case presentations, against a synthetic electronic health record cohort. Under progressive disclosure, 79.9% of simulated patients fell below the small-cell threshold ( k < 5 ) by the end of their disclosure sequence, with a median of seven disclosure steps to reach this threshold; when rare attributes were disclosed first, the median decreased to four steps. Even when no individual step disclosed a direct identifier, the cumulative quasi-identifier profile degraded k-anonymity below accepted safety thresholds. These findings suggest an operational limitation in per-prompt de-identification as applied to multi-turn clinical AI and large language model conversations: although HIPAA Safe Harbor includes a provision requiring no actual knowledge that remaining information could identify an individual, clinicians lack tools to assess cumulative re-identification risk in real time.
Schlagwörter
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Graduate School of Biomedical Sciences, The University of Texas Medical Branch at Galveston, Galveston, TX, United States.
- Department of Pathology, The University of Texas Medical Branch at Galveston, Galveston, TX, United States.
- Department of Computer Science, University of Pretoria, Pretoria, South Africa.
- Department of Pharmacology & Toxicology, The University of Texas Medical Branch at Galveston, Galveston, TX, United States.
- Transplant Division, Department of Surgery, The University of Texas Medical Branch at Galveston, Galveston, TX, United States.
Zitieren
APA
Weatherhead, J., Hasan, A., Weatherhead, J., Golovko, G., Grant, B., Garcia, J. D., Certuche, H. S., Powell, R. P., Abril, J. M., & McCaffrey, P. (2026). K-anonymity decay in multi-turn clinical large language model conversations. Frontiers in Digital Health, 8. https://doi.org/10.3389/fdgth.2026.1832168
Vancouver
Weatherhead J, Hasan A, Weatherhead J, Golovko G, Grant B, Garcia JD, et al. K-anonymity decay in multi-turn clinical large language model conversations. Frontiers in Digital Health. 2026;8. doi:10.3389/fdgth.2026.1832168
BibTeX
@article{Weatherhead2026,
author = {Weatherhead, James and Hasan, Azra and Weatherhead, Jake and Golovko, George and Grant, Bradley and Garcia, Juan David and Certuche, Hugo Sebastian and Powell, Reuben Peter and Abril, Jose Marri and McCaffrey, Peter},
title = {K-anonymity decay in multi-turn clinical large language model conversations},
journal = {Frontiers in Digital Health},
year = {2026},
volume = {8},
doi = {10.3389/fdgth.2026.1832168}
}
EUVIMED