Behavioral Privacy Leakage in Agentic Negotiation: Formalizing and Mitigating Inference Attacks via Randomized Policies
The paper targets what negotiation agents reveal through behavior, even when their stated constraints stay hidden.
Barkha Rani formalizes “behavioral privacy leakage” from concession paths, timing, and convergence patterns in multi-round agent negotiations. The proposed randomized negotiation policy is designed to provide differential privacy while still converging when agreement is possible. In 3,000 synthetic bilateral negotiations, it cut adversarial inference accuracy by 43–50% while keeping success rate and utility above 90%. The paper was accepted at the AI4TCI Workshop co-located with ARES 2026. HF Daily Papers' note
Barkha Rani formalizes “behavioral privacy leakage” from concession paths, timing, and convergence patterns in multi-round agent negotiations. The proposed randomized negotiation policy is designed to provide differential privacy while still converging when agreement is possible. In 3,000 synthetic bilateral negotiations, it cut adversarial inference accuracy by 43–50% while keeping success rate and utility above 90%. The paper was accepted at the AI4TCI Workshop co-located with ARES 2026. HF Daily Papers' note
score 4