Defense effectiveness across architectural layers: a mechanistic evaluation of persistent memory attacks on stateful LLM agents

TL;DR

This study systematically evaluates defenses against persistent memory attacks on open-source LLMs, revealing that most defenses fail except for memory-layer tool-gating, which can effectively prevent such attacks.

Contribution

It provides the first comprehensive analysis of defense effectiveness across architectural layers against persistent memory attacks on LLMs.

Findings

Most defenses fail to reduce attack success rates significantly.

Memory sandbox at the memory layer can reduce attack success to 0%.

Some defenses inadvertently invert their effectiveness depending on the model.

Abstract

Persistent memory attacks against LLM agents achieve high attack success rates against open-source models. In these attacks, malicious instructions injected via RAG-retrieved documents are stored in persistent memory and executed in later sessions. However, no systematic evaluation of defense effectiveness against this attack class exists. We evaluate six defenses across four architectural layers against delayed-trigger attacks on nine open-source models (5,040 runs, N=40 per condition). Four defenses fail at approximately baseline attack success rate: input-level filtering (Minimizer, Sanitizer) and retrieval-level filtering (RAG Sanitizer, RAG LLM Judge) achieve 88-89% ASR, statistically indistinguishable from the undefended baseline of 88.6%. Prompt Hardening partially fails at 77.8% ASR, with the reduction driven by two models at 0%: one genuine defense effect and one model-level refusal independent of the defense. The architectural explanation holds: input-level defenses cannot observe RAG-injected content, and retrieval-level classifiers are defeated by compliance-framed semantic masking. One defense, tool-gating at the memory layer (Memory Sandbox), reduces ASR to 0% for eight of nine models by removing the recall capability the attack requires. The exception inverts the defense entirely: a reasoning model that achieves 0% ASR under no defense via execution refusal inverts to 100% ASR under Memory Sandbox, because removing explicit recall forces the model onto the RAG pathway where its refusal mechanism does not activate. Memory Sandbox imposes zero utility cost in the absence of attack (BTCR = 100% across all conditions). These results provide the first systematic characterization of why each defense class fails against persistent memory attacks, enabling informed defense investment decisions.