Behavioral Privacy Leakage in Agentic Negotiation: Formalizing and Mitigating Inference Attacks via Randomized Policies

Autonomous negotiation agents risk leaking private constraints through observable concession trajectories and timing. This paper formalizes these inference attacks and introduces randomized policies to mitigate behavioral privacy leakage without compromising negotiation efficiency. The method aims to protect sensitive agent parameters in high-stakes procurement and insurance settings.