← Archive · 6G Tracker
 

Daily Dose of 6G

Edition 16 6 September 2026 Series: The architecture study

The acronym in the AI folder

The last of three editions on TR 23.801-01 goes to the key issue that asks an operator’s core to carry somebody else’s software agents. Its drafting folder holds twenty-seven files timestamped across the ten days to 28 August, and the word that keeps appearing in them is interim. The other thing in that work task is three letters no 3GPP source expands.

Transcribed from 3GPP’s published Release 21 timeline (© 3GPP 2026). The highlight holds where Editions 14 and 15 put it: TR 23.801-01 is a Stage 2 document, and Rel-21 SA Stage 2 complete is the milestone its conclusions have to reach.

The key issue that carries other people’s agents

Edition 15 took KI#18, AI for 6G Architecture — the network using AI on itself. Its twin under the same work task is the harder one. KI#19, 6G Network for AI, asks whether agents running on handsets should discover other agents through the 6G network, how they then communicate, what the network exposes to agents living on application functions, and whether the core itself should offer inferencing and training to applications.

That is the hinge. A network built for KI#19 has customers that are not people and not handsets but processes — addressable, authorisable, mobile between a phone, an edge site and a data centre. The document carrying the question is TR 23.801-01, “Study on Architecture for 6G System; Stage 2”, specification rapporteur Stefan Rommer of Ericsson LM, work-item rapporteur Malla Reddy Sama of NTT DOCOMO, at v0.8.0 of 28 July 2026.

Twenty-seven files, and the word “interim”

SA2’s open drafting area now shows all twenty-four key-issue folders last modified between 23 and 28 August 2026 — the week of SA2#176, Prague, 24–28 August. KI#19’s holds twenty-seven files, timestamped 19 to 28 August, and its file names carry a vocabulary Edition 15 did not see.

VariantWhat the file names call it
#19.1“Solution Variant #19.1 Update (Option1)”, carried from S2-2607830 into S2-2608469 and again into S2-2608700 on 26 August
#19.2An “interim conclusion for 19.2”, and a second file named for 19.1 and 19.2 together, both filed 25 August
#19.3“Update of network capability exposure” — four revisions between 19 and 27 August, plus a separate interim conclusion
#19.4“Baseline for AI model training” and a “Baseline paper for AI inference”, with “19.4 interim agreements” reaching r01 on 27 August

Two of those land on clauses of the key issue’s own objective, and not by inference — the file names say it. Variant 19.3 is the capability-exposure question; variant 19.4 is training and inference as something the network offers, and it is the busiest thread in the folder.

The new word is interim. Interim conclusions and interim agreements are how SA2 converges — the mechanism by which a key issue stops accumulating solutions and starts discarding them. Edition 15 could report activity in KI#18 and nothing more. Here the activity is pointed at closure. It is still only pointed: a file titled “interim agreements” is a proposal for interim agreements, and none of these documents was opened.

ACN: two expansions, neither of them 3GPP’s

In the SA2#170 archive from Gothenburg, August 2025, the WT#3_AI folder holds a file called S2-2506360r01-6G WT and KI proposal-ACN-v15 Merge for Net for AI - v3.3.docx. A merge document, in the AI work task, for what became Net for AI — KI#19. It is a Word file and returns 403.

Two expansions exist elsewhere, and they are not the same words. Huawei announced an ACN at MWC Barcelona on 1 March 2026 with three capabilities — digital identity management, dynamic group communication, collaboration task session management — describing it as reaching “beyond traditional SIM-based user identity management to support digital identity authentication and attribute registration for agents”. The headline calls it an Agentic Communication Network; the first paragraph calls it an Agent Communication Network.

The other belongs to Xiaodong Duan, Zhenglei Huang, Shiyu Liang, Shaowen Zheng, Lu Lu and Tao Sun, all of China Mobile Research Institute, in Frontiers of Information Technology & Electronic Engineering, 22 December 2025.

An AI-agent communication network: “a new paradigm to enable global information interaction and on-demand capability provisioning for single or multiple AI agents.”

Tao Sun is the TSG SA Vice Chair whose 3gpp.org article this series has quoted twice. A real overlap of name and employer, and no evidence at all about what the SA2 file means. A lead, printed as one.

Three letters this edition declines to expand

This newsletter does not know what ACN stands for in that file, and neither expansion above should be read across to it.

The source document was not opened. ACN appears in a file name in a public 3GPP folder. The file is .docx and refuses to serve, as every Word document on the server has all series. No 3GPP page, portal record or Highlights article fetched in sixteen editions expands the abbreviation.

The two external expansions differ from each other, and one source differs from itself. Huawei’s page prints Agentic in the headline and Agent in the body; the China Mobile paper says AI-agent. Neither is a document SA2 is obliged to agree with.

On Pen-holder Drafts. Edition 15 printed that path because a search index showed it. Today’s live listing of FS_6G_ARC returns twenty-seven entries — General, TR, one Rel-20 item and KI#1 through KI#24 — with no such folder among them. Both observations stand as recorded: a search index is not a directory, and neither cancels the other.

The meeting attribution is again ours. No folder here is labelled with a meeting; the timestamps simply fall around SA2#176’s dates. And the key-issue objective quoted above is from v0.4.0, 27 February 2026, through a third-party mirror, where the portal is at v0.8.0 of 28 July. Scope text is the stable part of an SA2 draft; solutions are not, which is why none is quoted.

Agent-centric is an argument, and it is being made outside the room

The framing has a literature. Mohamed Amine Ferrag, Abderrahmane Lakas and Merouane Debbah posted “6G Needs Agents: Toward Agentic AI-Native Networks for Autonomous Intelligence” on 2 May 2026, placing language-model agents in a semantic control plane “layered above deterministic 3GPP infrastructure”. That is a preprint, and its vocabulary is its own; none of it is 3GPP wording, and this series keeps the two apart deliberately.

What 3GPP has is a chain. Abdessamad El Moatamid and Xuefei Zhang of Huawei and Zhenning Huang of China Mobile, CT WG3 delegates, set it out on 3gpp.org: SA1 requirements, SA2 architecture, then CT3’s Stage 3 study FS_6G_AIP_CT, which must handle “communications between AI Agents on UEs via the 6G Network” and is weighing A2A, ANP, AGNTCY and MCP rather than inventing a protocol. KI#19 decides what CT3 is handed.

Three meetings to turn interim into conclusion

Work item FS_6G_ARC ends on 3 March 2027. After Prague, three SA2 meetings remain — October, Calgary in November, and South Korea in February 2027, ending five days before the deadline. Every interim agreement in that folder has to become a conclusion inside those three, or the question goes to Release 21 unanswered. On the evidence of the file names, KI#19 started early. On the evidence of an acronym nobody has expanded in public, it has not finished agreeing what it is building.

Sources

Daily Dose of 6G — tracking 3GPP Release 20's 6G study phase, one study at a time. Timeline after 3GPP's published Release 21 schedule.