about
MemoryDojo: Benchmarking Poisoning Through LLM Agent Memory (arxiv.org)
1 point by anatolii1309 7 days ago | hide | past | pdf | discuss on HN

In plain words: A test suite follows fake instructions planted in one session through an AI assistant's memory to see if they change its actions in later sessions. Every memory system tested was fooled, though by very different amounts, and tuning the planted text made attacks stronger.

Abstract

LLM agents benefit from using memory systems to improve performance on long-horizon, multi-session tasks. However, memory introduces a critical attack surface. Malicious instructions encountered in one session may be stored and propagated through memory before being retrieved and acted upon in later sessions. Because memory systems differ in how they process and retrieve information, the downstream effects of poisoning can vary significantly across systems. To quantify the effects of poisoning across different memory architectures, we introduce MemoryDojo, a benchmark for tracing the lifecycle of poisoned information through agent memory and measuring its influence on subsequent agent behavior. MemoryDojo is easily extensible, consists of 155 multi-session long-horizon trajectories covering four domains, and supports five different memory systems. Through our evaluation, we find that all memory systems are susceptible to memory poisoning, with substantial variation in downstream effects. We further propose a memory-aware poison optimization process that increases the poisoning efficacy, highlighting differences between memory architectures. We hope that MemoryDojo will support research on understanding and mitigating agentic memory risks.

David H. Yang, Blake Bullwinkel, Amanda Minnich, Mohammad Mohammadi Amiri, Giorgio Severi
arXiv:2611.09721 · cs.CR, cs.AI, cs.CL · submitted Nov 6, 2026
abstract · pdf · html

add comment on HN