Enhancing Data Privacy in Large Language Models Through Private Association Editing
Journal of Cybersecurity and Privacy · 01.06.2026 · 6 (3) · 96
DOI: 10.3390/jcp6030096 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Cybersecurity and Privacy, online erschienen am 01.06.2026. Verfasst von Davide Venditti und 6 weiteren Autor:innen. Die Arbeit nennt 58 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Fondo Italiano per le Scienze Applicate.
Abstract (Originalsprache)
Large language models (LLMs) require a significant redesign in solutions to preserve privacy in data-intensive applications due to their text-generation capabilities. Indeed, LLMs tend to memorize and emit private information when maliciously prompted. In this paper, we introduce Private Association Editing (PAE) as a novel defense approach for private data leakage. PAE is designed to effectively remove Personally Identifiable Information (PII) without retraining the model. We experimented on three open-weight, open-data models—GPT-Neo 1.3B, GPT-Neo 2.7B, and GPT-J—by applying Training Data Extraction (TDE) attacks to retrieve hundreds of PII, including email addresses, phone numbers, and Twitter handles. Since these models were trained on Pile, an openly available pre-training dataset, it is possible to verify the true extent of the data leakage. While all three models tend to leak PII under TDE attacks, experimental results demonstrate the effectiveness of PAE with respect to alternative baseline methods in defending against those attacks. In fact, unlike other techniques that tend to degrade model performance, our experiments show that PAE consistently reduces the number of leakages without affecting the model’s utility. We believe PAE will serve as a practical tool for removing memorized PII from deployed LLMs without retraining.
Förderung
- Fondo Italiano per le Scienze Applicate
Zitieren
APA
Venditti, D., Ruzzetti, E. S., Xompero, G. A., Giannone, C., Favalli, A., Romagnoli, R., & Zanzotto, F. M. (2026). Enhancing Data Privacy in Large Language Models Through Private Association Editing. Journal of Cybersecurity and Privacy, 6(3), 96. https://doi.org/10.3390/jcp6030096
Vancouver
Venditti D, Ruzzetti ES, Xompero GA, Giannone C, Favalli A, Romagnoli R, et al. Enhancing Data Privacy in Large Language Models Through Private Association Editing. Journal of Cybersecurity and Privacy. 2026;6(3):96. doi:10.3390/jcp6030096
BibTeX
@article{Venditti2026,
author = {Venditti, Davide and Ruzzetti, Elena Sofia and Xompero, Giancarlo A. and Giannone, Cristina and Favalli, Andrea and Romagnoli, Raniero and Zanzotto, Fabio Massimo},
title = {Enhancing Data Privacy in Large Language Models Through Private Association Editing},
journal = {Journal of Cybersecurity and Privacy},
year = {2026},
volume = {6},
number = {3},
pages = {96},
doi = {10.3390/jcp6030096}
}
EUVIMED