Understanding Stage-Wise Utility-Risk Trade-offs in LLM Agent Memory

Abstract

Long-term memory is becoming a core capability of LLM agents, enabling personalization and long-horizon interaction. However, memory mechanisms that retain, transform, or expose more information can affect both benign utility and susceptibility to memory poisoning. Existing evaluations typically measure memory utility or attack risk in isolation under fixed configurations, providing limited insight into how stage-specific design choices reshape their trade-off. We present \textsc{MemGauge}, a co

Categories

Framework mappings

Suggested from the entry's categories.

@misc{zang2026understanding,
  title = {{Understanding Stage-Wise Utility-Risk Trade-offs in LLM Agent Memory}},
  author = {Chuanchao Zang and Zi-Jian Cao and Xiangtao Meng and Jianing Wang and Wenyu Chen and Xinyu Gao and Li Wang and Zheng Li and Shanqing Guo},
  year = {2026},
  month = aug,
  eprint = {2608.30177},
  archivePrefix = {arXiv},
  url = {https://www.semanticscholar.org/paper/d9a499f524390e5452dad9a4818c8988506c1a65}
}