Mitigating Privacy Leakage in Agentic Negotiation via Randomized Policies

Negotiation agents risk leaking private constraints through observable concession trajectories and timing patterns. This method implements an adaptive stochastic policy that provides (epsilon, delta)-differential privacy while maintaining convergence to an agreement and maximizing utility.