Research
ICML Poster An Empirical Study of Memory Poisoning Defenses for LLM Agents
Publication date not yet evaluated · Added icml.cc
Page published
Coverage timeline
Single-source research — one report is available.
Why it matters
Memory poisoning of LLM agents can persistently corrupt an agent's stored context and drive malicious behavior, so empirical evaluation of defenses gives defenders evidence about what mitigations actually work.
An ICML 2026 poster titled "An Empirical Study of Memory Poisoning Defenses for LLM Agents" presents research evaluating defenses against memory poisoning attacks on LLM agents, with an associated code artifact (AMemGuard on GitHub).