arxivcs.AIcs.ET2026-06-27
Memory as an Attack Surface in LLM Agents: A Study on Multiple-Choice Question Answering
Shahnewaz Karim Sakib, Anindya Bijoy Das
AI agents extend conventional large language model (LLM) applications by integrating language understanding with task execution, external tool use, and memory mechanisms. While memory allows agents to retain prior interactions and provide more personalized and context-aware respo…