A privacy model for agentic negotiation

Barkha Rani · hf · 2026-07-21

This paper studies a subtler privacy problem in negotiation agents: even if private constraints are never explicitly revealed, an adversary can infer them from concession timing, trajectories, and convergence patterns. The authors formalize this as behavioral privacy leakage and propose an adaptive stochastic negotiation policy that aims to guarantee $(\varepsilon, \delta)$-differential privacy, almost-sure convergence when agreement is possible, and strong utility. On 3,000 synthetic bilateral negotiations, the mechanism cuts adversarial inference accuracy by 43–50% while keeping success rate and utility above 90%.

Original post →

More from Research

Research channel →