MemHarm is introduced, which predeclares a finite class of sparse, grounded semantic edits, evaluates candidates through the normal agent memory interface using offline paired-loss feedback, and certifies resolved selections within that class of sparse, grounded semantic edits.
Abstract
As LLM agents increasingly rely on persistent memory for long-horizon and personalized behavior, they can retain and reuse information across interactions, but this also creates a lasting channel through which malicious memory writes can influence future behavior. Persistent-memory attacks are typically evaluated by whether they succeed, yet successful attacks can leave persistent states with substantially different downstream consequences. We study this severity as a distinct attack-design objective and formalize it with counterfactual memory regret (CMR), the paired increase in expected downstream loss relative to clean memory. We introduce MemHarm, which predeclares a finite class of sparse, grounded semantic edits, evaluates candidates through the normal agent memory interface using offline paired-loss feedback, and certifies resolved selections within that class. Compared with attack-success optimization, CMR-guided selection produces substantially larger downstream loss while retaining most of the success-rate gain. Across two agent benchmarks and diverse memory designs, MemHarm attains the highest CMR point estimates among the evaluated general attacks on identical support. Factor-removal interventions link this harm to the selected semantic factor, and native-agent deployments verify the write-to-fresh-process attack path.
This work investigates whether a misaligned agent can write a goal it cannot yet act on to persistent memory, so that a future aligned agent carries it out when the opportunity arises, and investigates self-propagation of misalignment, which can propagate misalignment to more capable models.
Debeshee Das, Jacqueline Tay, Bruce Tsai et al.· 2 citations
A targeted prompt-level defense is evaluated and finds that it can reduce memory injection in many settings, but provides limited protection once the persistent memory has been poisoned.
Shu-Huai Huang, Jing-Feng Zhang, Hong Jia· 3 citations
This work proposes InjecMEM, a novel memory injection attack paradigm that requires only a single interaction to steer later responses of related queries toward a pre-specified output and achieves reliable topic-conditioned retrieval and targeted generation.
Han-Ling Tian, Gengyu Zhang, Zeyang Sha et al.· 7 citations
Control evaluations reveal that targeted poisoning risk varies across memory operations and motivate stage-aware evaluation and control of LLM-agent memory, showing that targeted poisoning risk varies across memory operations.
Chuan-Chao Zang, Zi-Jian Cao, Xiang-Tao Meng et al.· 0 citations
Multi-turn attacks on agentic systems can compose individually permissible actions into harmful outcomes, challenging defenses that assess actions or states in isolation. We show that such attacks leave a detectable signature in the agent's internal representations: harmful behavior emerges as an accumulated representa...
Hao-Yu Wang, Wei Zhao, Ye-Di Zhang et al.· 1 citation