Reverse-Engineering Model Editing on Language Models

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Sun, Zhiyu, Luo, Minrui, Wang, Yu, Chen, Zhili, He, Tianxing
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866916023285841920
author Sun, Zhiyu
Luo, Minrui
Wang, Yu
Chen, Zhili
He, Tianxing
author_facet Sun, Zhiyu
Luo, Minrui
Wang, Yu
Chen, Zhili
He, Tianxing
contents Large language models (LLMs) are pretrained on corpora containing trillions of tokens and, therefore, inevitably memorize sensitive information. Locate-then-edit methods, as a mainstream paradigm of model editing, offer a promising solution by modifying model parameters without retraining. However, in this work, we reveal a critical vulnerability of this paradigm: the parameter updates inadvertently serve as a side channel, enabling attackers to recover the edited data. We propose a two-stage reverse-engineering attack named \textit{KSTER} (\textbf{K}ey\textbf{S}paceRecons\textbf{T}ruction-then-\textbf{E}ntropy\textbf{R}eduction) that leverages the low-rank structure of these updates. First, we theoretically show that the row space of the update matrix encodes a ``fingerprint" of the edited subjects, enabling accurate subject recovery via spectral analysis. Second, we introduce an entropy-based prompt recovery attack that reconstructs the semantic context of the edit. Extensive experiments on multiple LLMs demonstrate that our attacks can recover edited data with high success rates. Furthermore, we propose \textit{subspace camouflage}, a defense strategy that obfuscates the update fingerprint with semantic decoys. This approach effectively mitigates reconstruction risks without compromising editing utility. Our code is available at https://github.com/reanatom/EditingAttack.
format Preprint
id arxiv_https___arxiv_org_abs_2602_10134
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Reverse-Engineering Model Editing on Language Models
Sun, Zhiyu
Luo, Minrui
Wang, Yu
Chen, Zhili
He, Tianxing
Cryptography and Security
Artificial Intelligence
Computation and Language
Large language models (LLMs) are pretrained on corpora containing trillions of tokens and, therefore, inevitably memorize sensitive information. Locate-then-edit methods, as a mainstream paradigm of model editing, offer a promising solution by modifying model parameters without retraining. However, in this work, we reveal a critical vulnerability of this paradigm: the parameter updates inadvertently serve as a side channel, enabling attackers to recover the edited data. We propose a two-stage reverse-engineering attack named \textit{KSTER} (\textbf{K}ey\textbf{S}paceRecons\textbf{T}ruction-then-\textbf{E}ntropy\textbf{R}eduction) that leverages the low-rank structure of these updates. First, we theoretically show that the row space of the update matrix encodes a ``fingerprint" of the edited subjects, enabling accurate subject recovery via spectral analysis. Second, we introduce an entropy-based prompt recovery attack that reconstructs the semantic context of the edit. Extensive experiments on multiple LLMs demonstrate that our attacks can recover edited data with high success rates. Furthermore, we propose \textit{subspace camouflage}, a defense strategy that obfuscates the update fingerprint with semantic decoys. This approach effectively mitigates reconstruction risks without compromising editing utility. Our code is available at https://github.com/reanatom/EditingAttack.
title Reverse-Engineering Model Editing on Language Models
topic Cryptography and Security
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2602.10134