Piyush Nagasubramaniam

Your Agent's Memories Are Not Its Own: Forged Reasoning Attacks on LLM Agent Memory and Defenses

Neeraj Karamchandani, Piyush Nagasubramaniam, Sencun Zhu, Dinghao Wu

Persistent memory has enabled large language model (LLM) agents to store factual knowledge, prior decisions, reasoning histories, tool usage information, and context. While this has improved the agent's functionality and continuity across tasks, it has also introduced a new attac…Read more