Bad Memory: Evaluating Prompt Injection Risks from Memory in Agentic Systems

Abstract

A growing class of agentic systems maintain persistent state across sessions through memory files, behavioral preferences, and knowledge bases. While this makes agents more useful and self-improving, it also creates a new attack surface for prompt injections in which malicious instructions can be embedded within persistent files and influence future behavior. In this work, we study prompt injection attacks in memory-based agentic systems using a sandboxed synthetic workspace. We evaluate two age

Categories

Framework mappings

Suggested from the entry's categories.

@misc{gadgil2026bad,
  title = {{Bad Memory: Evaluating Prompt Injection Risks from Memory in Agentic Systems}},
  author = {Soham Gadgil and David Alexander and Sai Sunku and Franziska Roesner},
  year = {2026},
  month = jul,
  eprint = {2607.14611},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2607.14611}
}