OpenANA2A-TLinux-Foundationopen-sourcekey
The A2A-T protocol Huawei proposed at MWC in March has reached open source. Linux Foundation Networking (LFN) announced OpenAN as a Candidate project, with China Mobile as chair and Huawei, ZTE, AsiaInfo, Orange, AIS, Infosys and Personal as founding members; the global launch ceremony was held at MWC Shanghai on 25 June. Huawei and China Mobile have jointly contributed the first code drop: an A2A-T SDK implementing the A2A-T extensions (message encapsulation and decapsulation), a Registry Center handling registration, discovery and full lifecycle management for both network-layer and service-layer agents, and an Orchestration Center for multi-agent workflow design and execution that can import TM Forum Autonomous Networks solution packages directly. The A2A-T extensions on top of plain A2A are four per the official architecture page: structured tasks (Task-T), network event subscription and publication (Notification-T), inter-agent arbitration (Negotiation-T) and security authorisation (Authorization-T). China Mobile has validated A2A-T in live network trials in Fujian: wireless work orders down 30 percent and mean time to repair cut from 3 hours to 2.03 hours. Why this matters for the agent gateway: the OpenAN registry and orchestration centres address the same problems as draft-zhang-dmsc-gateway-directory-sync (gateway capability directory) and draft-han-rtgwg-agent-gateway-intercomm-framework (gateway interconnection) on the IETF side, but the open source stack is already further along. The code lives under the project-openan organisation on GitHub (mirrored on AtomGit under the OpenAN name): the Registry Center (Python and PostgreSQL, full AgentCard lifecycle management and semantic search) and the Orchestration Center (React and FastAPI, drag-and-drop workflow design with A2A-T negotiation) are publicly readable, and the A2A-T SDK ships in both Python and Java. All components are Apache-2.0 licensed. Source: LF Networking announcement (2026-06-24) and the OpenAN website.
A2A-THuaweiautonomous-networksmulti-agent-protocolkey
At the FutureNet World summit in London on 22 April 2026, Huawei laid out four priority tasks for the industry alongside its upgraded ADN L4 solution. The fourth goes straight at the protocol layer: jointly define and extend A2A-T so operators can coordinate efficiently across domains and vendors, making it the preferred implementation path for autonomous-network scenario packages and reshaping how cross-layer integration is done. Huawei says it will release A2A-T as an open source project during the TM Forum DTW summit in June and submit it to the Linux Foundation. Why it matters for the agent gateway: this is the first multi-agent protocol proposed by a network vendor explicitly for cross-domain operator coordination, and it tackles the same problem as draft-han-rtgwg-agent-gateway-intercomm-framework on the IETF side. The two paths are worth tracking together. Source: Huawei carrier BG (2026-04-22).
A2A-THuaweiChina-MobileGLOMOautonomous-networks
The autonomous-network "AI+Network" operations transformation built jointly by Huawei and China Mobile won the GSMA GLOMO award for Best AI-Enabled Network Solution. In its release Huawei states plainly that the solution originates the first carrier-grade A2A-T multi-agent communication protocol, enabling autonomous coordination across domains and lifting network autonomy from single-point intelligence to collective intelligence. The solution is in commercial use across 5GtoC, 5GtoV and 5GtoB; in 5GtoC it automates diagnosis for over 90 percent of alarms and cuts mean time to repair by 25 percent. The "carrier-grade" framing is the part worth noting: it pulls multi-agent protocols out of the application layer, where A2A and MCP live, and into an operator network frame. Source: Huawei (2026-03-04).
ATHCAICTtrusted-handshakeopen-sourcetrust-layer
On 29 April 2026, the China Academy of Information and Communications Technology (CAICT) launched version 1.0 of the Agent Trust Handshake protocol together with China Telecom, China Mobile, China Unicom, Huawei, Tencent, ZTE and the Chinese University of Hong Kong, Shenzhen. The code is public on AtomGit and GitHub. ATH sits between user interaction and the communication protocols, filling the trust layer that MCP and A2A leave open: agent identity and verification, trusted handshake and session establishment, auditing and trust evaluation. It uses a nine-step handshake across three roles (user, agent, application) and follows user sovereignty, least privilege and end-to-end traceability. CAICT is explicit that ATH complements rather than competes with MCP and A2A. The first ATH capability assessment, published on 8 July 2026, qualified 13 vendors, including the agent security gateway from China Telecom Research Institute. Source: ATH repository on AtomGit.
wechatChina MobileAONPlaunch
China Mobile unveiled the Agent Open Network Protocol (AONP) framework together with an agent gateway at MWC 2026. AONP has previously appeared on the IETF side as an individual draft; this is the operator-side product launch, driven by the observation that hundreds of millions of AI agents now interoperate over the public Internet and that the network layer needs an open protocol and matching gateway capability. For anyone tracking the agent gateway, this is the clearest signal yet that a Chinese operator treats it as a network element rather than an application-side component. Source: WeChat public account "中移智库" (2026-03-04).
wechatChina TelecomIETFprototype
China Telecom Research Institute demonstrated a self-developed agent gateway at the Hackathon of the 125th IETF meeting, achieving intent-driven cross-domain agent coordination. Unlike the draft-stage work, this is a running prototype shown publicly at a standards meeting — evidence that the agent gateway has moved from concept to something demonstrable. Two of the three Chinese operators have now shown their hand within six months, which makes this worth tracking as a standing industry thread. Source: WeChat public account "科创电信" (2026-03-16).
wechatIoAtechnology-analysis
A breakdown of the key technologies behind the Internet of Agents: bearer network technologies, agent gateway and interconnection protocol suites, identity trust and lifecycle management, intent routing and capability discovery, and distributed multi-agent coordination. The taxonomy closely mirrors what RTGWG and DMSC drafts are converging on, which makes it a useful Chinese-language companion read — a way to read the English drafts back into their domestic industry context. Source: WeChat public account "鹏博士研究院" (2026-08-28).
wechatF5AI-gatewayproduct
F5 released a next-generation AI gateway aimed at agents, applying security and governance policy to every AI request and providing a single control plane for access to models, agents and tools. Worth reading against the standards work: the vendor notion of an "AI gateway" is about in-enterprise governance and cost control, whereas the standards bodies are debating an "agent gateway" for cross-domain interconnection. The two are not the same thing, and holding them side by side makes the boundary much easier to see. Source: WeChat public account "F5 Inc" (2026-08-26).
OpenANA2A-TSDKTM-Forumopen-sourcekey
The project-openan organisation on GitHub hosts thirteen repositories, including assets absent from domestic coverage: a2a-t-sdk-python (about 4,600 lines, v1.0.0, Alpha status), a2a-t-sdk-java, a2a-t-prompt-templates-spec (the prompt template specification) and workflow engine SDKs. Three hard conclusions from reading the SDK source. First, the actual Task-T wire format is minimal — a YAML front matter of three fields (scenario_code / language / description) plus a free-form body, so the much-touted "deterministic task schema" lands on prompt engineering rather than a new message protocol. Second, extension declarations are uniformly anchored in the TM Forum namespace projects.tmforum.org/a2aproject/telecommunication/extensions/ with v1 versioning, which makes TM Forum — not the IETF — the home ground for A2A-T standardisation. Third, the SDK documentation is explicit that it "does not handle AgentCard registration, authentication, service hosting or business execution"; the official architecture page lists four extensions (Task-T / Notification-T / Negotiation-T / Authorization-T), yet SDK v1.0.0 implements only Task-T and Negotiation-T (three negotiation flows: information / feasibility / target), Notification-T appears only in documentation examples, and Authorization-T scores zero hits across the entire codebase — the security extension is still on paper, which corroborates the registry center approach of blacklisting text instead of parsing extensions. Source: OpenAN website architecture page and the project-openan organisation on GitHub.
10 items Agent Gateway Focus · 2026-08-31
group-standardISCpublic-commentaction
The Internet Society of China has released drafts of two group standards for public comment — "Internet of Agents: Networked Agent Gateway Capability Requirements" and "Internet of Agents: Task Orchestration — OpenClaw-like Multi-Agent Collaboration Platform Capability Requirements". Feedback closes 18 September 2026 via the standard comment form. This is the first Chinese group standard to target the networked agent gateway as a standardisation object in its own right. Source: Internet Society of China (2026-08-19).
group-standardISCdata-agentcall-for-participants
CAICT is leading development of the group standard "AI — Data Agent General Requirements", to be registered with the Internet Society of China; contributing organisations are being recruited. The standard will define the basic definition, overall architecture, core capabilities and generic requirements for data agents, and is intended to interlock with existing work on agent interconnection and tool invocation. Source: Internet Society of China (2026-08-28).
group-standardISCregistration
The Internet Society of China announced 11 newly registered group standards. Directly agent-related is a "Data Agent Service Capability Maturity Model"; also registered are an AI-enhanced DevOps maturity model (platform requirements and continuous delivery parts), AIOps observability for banking, and a risk assessment method for cross-border transfer of AI training data. Source: Internet Society of China (2026-08-28).
IETFRTGWGrequirementsagent-gateway
Sets out the requirements for introducing Agent Gateways into agent-to-agent communication, and analyses why existing hardware and software gateways fall short — on scalability, communication efficiency and security — arguing that the Agent Gateway must be defined as a new class of network entity. Authors: Bing Liu, Nan Geng, Xiaotong Shang, Qiangzhou Gao, Zhenbin Li, Jing Gao. Source: IETF Datatracker (2026-05-31).
IETFRTGWGframeworkinterconnection
Defines the framework and requirements for intercommunication between Agent Gateways (AGw) in the Internet of Agents, specifying a hierarchical layered model, functional components, protocol requirements and deployment considerations. It targets data synchronisation, protocol compatibility and security in cross-domain deployments. At revision 02, it is the most fully developed of this batch. Authors: Han Zhengxin, Zheng Ruan, Mengyao Han, Jinjie Yan, Bing Liu, Tao He, Ran Pang. Source: IETF Datatracker (2026-07-05).
IETFOPSAWGpolicy-controloperator
Defines an operational policy control model for operator-managed Agent Gateways: how an operator controls and observes interactions that the gateway admits, mediates, routes, proxies or otherwise handles. The draft draws an explicit boundary — it does not govern an agent's internal behaviour, reasoning, planning, prompts, memory, tools, runtime or lifecycle. That separation matters a great deal for operator deployment. Authors: Jing Zhao, Ran Pang, Shuai Zhang. Source: IETF Datatracker (2026-07-06).
IETFarchitecturesemantic-routingworking-memory
Presents an architectural framework for an Agent Communication Gateway (Agent-GW) supporting large-scale, heterogeneous, dynamic multi-agent collaboration across administrative and protocol boundaries. It argues the infrastructure must move from rigid endpoint-based connectivity to intent-based communication, and introduces the notion of working memory. Authors: Xiaohui Xie, Wang Zian, Tianshuo Hu, Yong Cui. Source: IETF Datatracker (2026-07-13).
IETFDMSCmediationsemantics
Proposes a semantic mediation layer on the gateway. The core argument: a cross-domain collaboration request may need to combine a stated goal, capability profiles, tenant or domain policy, trust evidence, disclosure limits, operational constraints and handoff context before a concrete interaction mechanism can be chosen — and those mediation decisions belong neither to the agent client nor to the agent server, only to the gateway. Source: IETF Datatracker (2026-07-02).
IETFDMSCmediationrevision-watch
The -00 initial draft from the same author team, covering the same gateway mediation layer; the abstract is currently near-identical to the semantic-layer draft. Two coexisting drafts are a normal early-stage pattern — expect consolidation or eventual withdrawal of one. Track revision timestamps rather than counting both as separate contributions. Source: IETF Datatracker (2026-07-05).
IETFDMSCcapability-directorycontrol-plane
Describes a gateway capability-directory framework for Internet of Agents deployments that use Agent Gateways. The central claim is that a gateway-managed capability directory is a necessary control-plane function for maintaining validated capability information, and that one cannot rely on transient advertisements, static endpoint bindings or external descriptions alone — an argument that runs parallel to the operator-side notion of an inventory or resource ledger. Authors: Lianhua Zhang, Huiling Yang, Yun Li, Shoufeng Wang. Source: IETF Datatracker (2026-04-29).
This brief was AI-assisted in collection and summarization, then human-reviewed before publication.