Zero-trust A2A communication

Control objective

Zero-trust A2A communication checks that when autonomous agents talk to each other (agent-to-agent messaging in a multi-agent banking system), every message is authenticated to a verified identity and authorised against policy before it is acted on, and that agent populations are kept in separate network segments. It matters because in agentic systems a compromised or impersonating agent can issue instructions to others — for example a rogue agent telling a payments agent to release funds — and without per-message verification these instructions are trusted blindly, the threat captured by OWASP ASI07, MITRE AML.T0051.002 and MAESTRO L3/L6. To implement, give each agent a strong cryptographic identity, verify the sender's identity on every inbound A2A message, run each message through a policy/authorisation check before execution, and place agent groups in segmented networks so a breach in one cannot freely reach others; log identity, policy decision and routing for every message as evidence. This control has no numeric metric or formula, so judge it against its stated threshold: 100% of A2A messages must be identity-verified and policy-checked, with network-segmented populations. A breach — any unverified or unauthorised message, or flat network topology — is a Critical-priority failure that must block the message and trigger isolation and review.