MemAudit: Post-hoc Auditing of Poisoned Agent Memory via Causal Attribution and Structural Anomaly Detection

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Tan, Zhewen, Yao, Yilun, Jin, Huiyan, Yu, Wenhan, Wang, Guoan, Fan, Mengyuan, lu, liang, Liu, Feng, Zhang, Xiangzheng, Ma, Duohe, Yang, Tong, Sun, Lin
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917524903297024
author Tan, Zhewen
Yao, Yilun
Jin, Huiyan
Yu, Wenhan
Wang, Guoan
Fan, Mengyuan
lu, liang
Liu, Feng
Zhang, Xiangzheng
Ma, Duohe
Yang, Tong
Sun, Lin
author_facet Tan, Zhewen
Yao, Yilun
Jin, Huiyan
Yu, Wenhan
Wang, Guoan
Fan, Mengyuan
lu, liang
Liu, Feng
Zhang, Xiangzheng
Ma, Duohe
Yang, Tong
Sun, Lin
contents Large language model agents increasingly rely on persistent memory to store past interactions, retrieve relevant demonstrations, and improve long-horizon task execution. However, this memory mechanism also creates a practical security vulnerability: an adversarial user may inject malicious records into the agent's memory through ordinary interaction, and these records can later be retrieved to steer the agent's reasoning and actions. Existing defenses primarily focus on online intervention, such as prompt filtering or output blocking, but they do not address the post-hoc question of which stored memories are responsible after harmful behavior has already been observed. We propose \textbf{MemAudit}, a post-hoc causal memory auditing framework for memory-augmented LLM agents. The framework combines two complementary signals: (1) a counterfactual memory influence score that measures each memory's causal contribution to harmful outputs, and (2) a memory consistency graph that identifies structurally anomalous memories within the broader memory store. We evaluate MemAudit against MINJA, a query-only memory injection attack in which malicious records are generated and stored through normal agent interactions rather than direct memory-bank modification. Across both QA and reasoning-agent settings, MemAudit substantially reduces attack success rates under realistic post-hoc auditing scenarios. The results show that QA attack success is reduced from $70\%$ to $0\%$, while RAP attack success drops from $83.3\%$ to $0\%$.
format Preprint
id arxiv_https___arxiv_org_abs_2605_23723
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle MemAudit: Post-hoc Auditing of Poisoned Agent Memory via Causal Attribution and Structural Anomaly Detection
Tan, Zhewen
Yao, Yilun
Jin, Huiyan
Yu, Wenhan
Wang, Guoan
Fan, Mengyuan
lu, liang
Liu, Feng
Zhang, Xiangzheng
Ma, Duohe
Yang, Tong
Sun, Lin
Artificial Intelligence
Large language model agents increasingly rely on persistent memory to store past interactions, retrieve relevant demonstrations, and improve long-horizon task execution. However, this memory mechanism also creates a practical security vulnerability: an adversarial user may inject malicious records into the agent's memory through ordinary interaction, and these records can later be retrieved to steer the agent's reasoning and actions. Existing defenses primarily focus on online intervention, such as prompt filtering or output blocking, but they do not address the post-hoc question of which stored memories are responsible after harmful behavior has already been observed. We propose \textbf{MemAudit}, a post-hoc causal memory auditing framework for memory-augmented LLM agents. The framework combines two complementary signals: (1) a counterfactual memory influence score that measures each memory's causal contribution to harmful outputs, and (2) a memory consistency graph that identifies structurally anomalous memories within the broader memory store. We evaluate MemAudit against MINJA, a query-only memory injection attack in which malicious records are generated and stored through normal agent interactions rather than direct memory-bank modification. Across both QA and reasoning-agent settings, MemAudit substantially reduces attack success rates under realistic post-hoc auditing scenarios. The results show that QA attack success is reduced from $70\%$ to $0\%$, while RAP attack success drops from $83.3\%$ to $0\%$.
title MemAudit: Post-hoc Auditing of Poisoned Agent Memory via Causal Attribution and Structural Anomaly Detection
topic Artificial Intelligence
url https://arxiv.org/abs/2605.23723