IETF Publishes Draft Protocol to Verify Trust Between AI Agents

The Internet Engineering Task Force (IETF) has published draft-sharif-attp-00, the Agent Trust Transport Protocol (ATTP), to address security vulnerabilities in AI agent-to-agent communication. The protocol introduces a five-dimension trust scoring model covering identity, reputation, and anomaly detection, replacing unreliable self-declared agent credentials. Current systems like the A2A protocol rely on agent cards — unverified, self-declared claims — which leave networks exposed to prompt injection, Sybil attacks, and compromised agents. ATTP uses cryptographic identity verification and historical behavioral data to derive trust scores, rather than accepting claims at face value. The protocol aims to prevent scenarios where a single malicious or compromised agent can manipulate decisions across an entire agent network.