Research
MemoryDojo: Benchmarking PoisoningThrough LLM Agent Memory
Publication date unknown · Discovered arxiv.org
Page published
Publication date unknown · First observed: 12 Nov 2026
Coverage timeline
Single-source research — one report is available.
Why it matters
MemoryDojo's findings highlight how agent memory can turn a transient prompt injection into persistent behavior, making cross-session testing important for defenders evaluating memory-enabled agents.
MemoryDojo is a research benchmark tracing how poisoned information passes through LLM-agent memory and influences later behavior, covering 155 multi-session trajectories across four domains and five memory systems. The authors report that all evaluated memory systems were susceptible to poisoning, with varying downstream effects, and that memory-aware payload optimization increased poisoning efficacy.