The Protocol Stack Nobody Audited: Three Agent Protocols and the Cross-Protocol Audit Gap

The three-protocol enterprise agent stack — MCP, A2A, and UCP — was designed by three teams with three threat models. (ACP, once listed as a fourth, merged into A2A under the Linux Foundation on 2025-08-29.) MCP lists structured audit trails as roadmap work; A2A obtains credentials out-of-band, outside its own scope; UCP signs only single transactions. Reading the three design documents together exposes a cross-protocol accountability void no single document owns. This analytical white paper traces a single request across all three layers, maps the regulatory exposure to NIST AI RMF, SR 26-2 (the 2026 successor to SR 11-7, applied by analogy), the EU AI Act, ISO 42001, HIPAA, and SOC 2, and sketches what a cross-protocol audit surface would require — with an explicit Honest Limits section, because no breach has been attributed to a cross-protocol failure and the core argument is inference, not incident.