EchoGuard: An Agentic Framework with Knowledge-Graph Memory for Detecting Manipulative Communication in Longitudinal Dialogue

ABSTRACT

Detecting deceptive or manipulative intent across long-term interactions is a major blind spot for standard stateless LLMs. The authors introduce EchoGuard, an agentic framework that utilizes a persistent knowledge-graph memory to track interpersonal dynamics and psychological manipulation over extended periods. By continually updating semantic sub-graphs of user interactions, this system achieves a new state-of-the-art in conversational threat detection.