An Apple research paper identifies behavioral privacy leakage in autonomous negotiation agents. Adversaries can infer private constraints by analyzing concession trajectories and timing, bypassing standard cryptographic protections. The team proposes randomized policies to mitigate these inference attacks. This provides a necessary security framework for agents deployed in high-stakes insurance and procurement settings.