Harry Potter is Still Here! Probing Knowledge Leakage in Targeted Unlearned Large Language Models via Automated Adversarial Prompting

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: To, Bang Trinh Tran, Le, Thai
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866918031782838272
author To, Bang Trinh Tran
Le, Thai
author_facet To, Bang Trinh Tran
Le, Thai
contents This work presents LURK (Latent UnleaRned Knowledge), a novel framework that probes for hidden retained knowledge in unlearned LLMs through adversarial suffix prompting. LURK automatically generates adversarial prompt suffixes designed to elicit residual knowledge about the Harry Potter domain, a commonly used benchmark for unlearning. Our experiments reveal that even models deemed successfully unlearned can leak idiosyncratic information under targeted adversarial conditions, highlighting critical limitations of current unlearning evaluation standards. By uncovering latent knowledge through indirect probing, LURK offers a more rigorous and diagnostic tool for assessing the robustness of unlearning algorithms. All code will be publicly available.
format Preprint
id arxiv_https___arxiv_org_abs_2505_17160
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Harry Potter is Still Here! Probing Knowledge Leakage in Targeted Unlearned Large Language Models via Automated Adversarial Prompting
To, Bang Trinh Tran
Le, Thai
Computation and Language
Artificial Intelligence
Cryptography and Security
Machine Learning
This work presents LURK (Latent UnleaRned Knowledge), a novel framework that probes for hidden retained knowledge in unlearned LLMs through adversarial suffix prompting. LURK automatically generates adversarial prompt suffixes designed to elicit residual knowledge about the Harry Potter domain, a commonly used benchmark for unlearning. Our experiments reveal that even models deemed successfully unlearned can leak idiosyncratic information under targeted adversarial conditions, highlighting critical limitations of current unlearning evaluation standards. By uncovering latent knowledge through indirect probing, LURK offers a more rigorous and diagnostic tool for assessing the robustness of unlearning algorithms. All code will be publicly available.
title Harry Potter is Still Here! Probing Knowledge Leakage in Targeted Unlearned Large Language Models via Automated Adversarial Prompting
topic Computation and Language
Artificial Intelligence
Cryptography and Security
Machine Learning
url https://arxiv.org/abs/2505.17160