What happened
On 30 October 2026 a new individual Internet-Draft, draft-manyfolks-agentproto-arch-00, was submitted (last updated 2026-10-30) by J. Rosenberg (jdrosen.net), Z. Sarker and T. Reddy (Nokia), K. Yao (China Mobile), D. Liu (Alibaba) and S. Krishnan (Cisco). It defines the terminology and architecture for AI agent communications protocols over the Internet and establishes the foundational architectural elements for the Agent Communications Protocol (ACP), also known as 'agentproto', a new protocol to be developed in the IETF. It specifies three interaction patterns — user-agent client to AI agent, AI agent to a session-oriented tool, and AI agent to AI agent — and argues these require a new session-like protocol that handles lifecycle management (creation, modification, deletion) and exchange of input/output across long-lived, intermittent, multi-host sessions. Discussion is under the IETF Agent Communication Protocols (agentproto) mailing list.
Why it matters
This is the first architecture framing a proposed IETF standard for AI-agent communications. Its session, identity and lifecycle model will shape how agent-to-agent, agent-to-tool and agent-to-user sessions are authenticated, authorized and managed at internet scale — the control plane for secure agentic AI. It signals an emerging IETF standards track for agent interoperability that security teams and agent-gateway vendors will need to align with.
Action needed
Follow the IETF Agent Communication Protocols WG; review the ACP architecture for security, identity and authorization implications and contribute to the protocol requirements/design work.