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

Researchers have identified a security vulnerability where autonomous negotiation agents can unintentionally reveal sensitive private data through their bargaining patterns. This study demonstrates that third parties can infer confidential information by analyzing an agent’s behavioral strategy during interactions. To address this risk, the authors propose a defense mechanism using randomized policies to mask behavioral signals. This research highlights the security challenges associated with deploying agentic AI in sensitive domains, providing a formal framework to mitigate inference attacks during automated negotiations.