🤖 AI Summary
While existing agent-to-agent (A2A) interoperability protocols encrypt message content, they inadvertently leak communication graph metadata, enabling adversaries to infer and compromise the integrity of incomplete workflows. This work elevates the threat model of metadata exposure from a privacy concern to a critical workflow integrity risk. We introduce revelatory characteristics centered on semantics, foresight, and execution coupling, and formally define metadata privacy properties at both the transport and orchestration layers. Building upon technologies such as SimpleX/SMP, Tor, and mix networks, we design a metadata-protecting transport binding compatible with A2A protocols. Experimental results demonstrate that raw metadata alone enables high-accuracy task-type classification; however, when our proposed privacy mechanisms are applied, classification performance degrades to random guessing, substantially diminishing adversarial advantage.
📝 Abstract
Agent-interoperability protocols such as A2A and MCP standardize what agents say to one another, but assume address-based transport over HTTP(S). Such transports protect message content, increasingly with end-to-end encryption. What they leave in the clear is the communication graph: which agent contacts which, when, and how often. In agent systems this graph is more consequential than a privacy framing suggests. Endpoints are often capability-labeled, workflows are structured and chained, and interactions are coupled to real actions, so an observer recovers more than past relationships. It can infer the pending workflow, the task being assembled and the action likely to follow. At machine speed, it can act on that inference before the workflow completes. The threat is therefore one of workflow integrity, not privacy alone: predictive leverage over autonomous action. We give a threat model for the agent communication graph; identify what makes agent metadata distinctively revealing (semanticity, prospectivity, actuation); define transport- and bootstrap-layer privacy properties and weigh candidate transports (SimpleX/SMP, Tor, mixnets) against them; and present an A2A case study in which a metadata-protecting binding is expressible but surfaces the protocol's identity assumptions. We test these on a generative model anchored to a real A2A capture. From passive metadata alone, with no payloads, a classifier recovers a task's class well above chance, from only the workflow's opening; applied together, the properties drive that recovery sharply back toward chance. Beyond what an observer can recover, we measure the leverage of acting on the leak: from a workflow's opening and under a fixed budget, an adversary choosing which workflows to act on realizes in this model most of a clairvoyant attacker's advantage over a metadata-blind one, and the same properties suppress it.