Behavioral Privateness Leakage in Agentic Negotiation: Formalizing and Mitigating Inference Assaults by way of Randomized Insurance policies

This paper was accepted on the AI4TCI (Workshop on AI for Safe and Reliable Essential Infrastructure Techniques) Workshop on the Worldwide Convention on Availability, Reliability and Safety (ARES) 2026.

Autonomous negotiation brokers are more and more deployed in high-stakes settings comparable to insurance coverage and procurement. Whereas cryptographic strategies shield explicitly disclosed constraint values, they fail to deal with a subtler risk: behavioral privateness leakage, the place an adversary infers non-public constraints from observable negotiation dynamics comparable to concession trajectories, timing, and convergence patterns. This paper investigates behavioral differential privateness in multi-round negotiation protocols. We design an adaptive stochastic negotiation coverage that collectively ensures (ε,δ)-differential privateness, almost-sure convergence of the supply sequence (reaching settlement when the counterparty’s reservation worth permits), and excessive negotiation utility. Evaluated on 3,000 artificial bilateral negotiations, our mechanism reduces adversarial inference accuracy by 43–50% whereas sustaining a negotiation success price and utility above 90%, demonstrating that robust privateness ensures may be achieved with out vital lack of efficiency.