SA2, the 3GPP working group that designs the core-network architecture, meets in Calgary from 16 to 20 November 2026, alongside 14 other working groups. The meeting carries a deadline. In September the SA plenary asked SA2 to decide the overall 6G system architecture "as soon as possible and no later than SA2#178 (November 2026)" and to report the outcome to TSG SA#114 in December (SP-260999). In Prague in October SA2 did not choose between three candidate architectures. It only agreed interim principles, as our summary of SA2#177 described. This preview is based on the documents submitted for Calgary, read before the meeting starts. Nothing in it has been decided yet.

The short version: the three options are no longer argued as three options. Companies have submitted "converged" proposals, and the rapporteurs (NTT DOCOMO and Ericsson) have merged them into one compromise text. The agreed floor is fairly broad: a service-based core that evolves from 5G, and essential connectivity that does not depend on AI. What remains open is narrower but still real: whether AI agents become standardized network functions, where NAS routing lives, and how a phone's intents reach the core. A second deadline concerns whether a 6G phone may keep its voice calls on 5G.

The compromise on the table

The rapporteurs' consolidated summary (S2-2612551) lists the overall-architecture inputs submitted for Calgary and proposes a new clause 8, "Consolidated 6G architecture interim agreements", for the study report TR 23.801-01. An intent is a request that states a goal rather than a procedure, for example a service level an application wants. The text proposes:

  • Common principles. The 6G core is a service-based architecture (SBA) with standardized network functions (NFs), and it builds on the evolution of the 5G core. Registration, mobility, connection and session management stay standardized end-to-end, and their execution is specified "without relying on the deployment or availability of AI functionality".
  • Intent handling and AI. 3GPP would standardize the functional responsibilities (processing intents, planning, coordinating, invoking network capabilities) and the interfaces other NFs see. How that is built inside would be left to implementation. AI agents would be one implementation option, and the architecture "does not mandate a specific decomposition into separate standardized AI Agent NFs".
  • How AI calls the network. Through standardized NF services. Semantic descriptions of those services, called "Tools" in some proposals, are left to the AI key issue (KI#18).
  • NAS routing. NAS is the signalling protocol between the phone and the core. The xAMF, the working name for the 6G access and mobility function, would be the single NAS entry point and would contain the routing function. "The architecture does not mandate a separate NAS routing/transport NF." Whether routing is separated from mobility management inside the xAMF is left to KI#1.1.
  • Intent transport. Both the control plane (NAS) and the user plane would be supported, and a phone that supports intents would have to support both. The network picks one. The summary itself notes concerns that this adds phone complexity and complicates roaming.

Most points come with an explicit question for SA2. If SA2 agrees, the text becomes an interim agreement in the study report, not a specification.

Who still pulls in which direction

  • Agent-based core. A proposal from 16 sources led by China Mobile, including China Unicom, China Telecom, CAICT, Huawei, ZTE, OPPO, vivo, Xiaomi, Honor and Toyota (S2-2612168), defines a System AI Agent NF and a Connection AI Agent NF. It enhances NF services with semantic descriptions so that agents can invoke them as Tools. It separates NAS routing from mobility management, with whether that leaves the xAMF still open, and carries phone intents over the control plane. Huawei's own proposal (S2-2612491) also builds on AI agents.
  • AI as an optional, separate part. Ericsson, Qualcomm, Rogers, Telstra, AT&T, Google and Federated Telecoms Hub (S2-2612541) make "Intent Handling and AI Services" an optional functional part. Only its external interfaces would be standardized, so that it can "evolve faster than 3GPP specified solutions normally are". Intents would go over the user plane only, and new NAS transport would be added only for new services.
  • A dedicated intent NF. MediaTek, Samsung, SK Telecom, LG Uplus, Sony, ETRI, Vodafone, NTT DOCOMO, Chunghwa Telecom, Lenovo and Ewha Womans University (S2-2611712) propose one dedicated NF for intents whose internal logic is not specified, structured intent expressions, and transport by dedicated NAS message or user plane.
  • Keep 5G NAS as it is. Vodafone, NTT DOCOMO and AT&T (S2-2612550) argue against a pure-routing NAS model. By their reading of TS 24.501 the best gain is 4 bytes saved per NAS transport message, while the change moves AMF functions to other NFs and needs a new roaming interface. They also argue that core changes with no visible end-user benefit have historically slowed radio rollouts, citing VoLTE and 5G standalone.

Two inputs aim at the floor rather than at an option. The operator principle that "essential connectivity procedures (such as registration, mobility, session management and QoS) shall not depend on AI functionality" was postponed in Prague. It is back as S2-2612277 with 18 sources instead of 9: AT&T, Boost Mobile, BT, Deutsche Telekom, EchoStar, NTT DOCOMO, T-Mobile USA, Verizon, Vodafone, KDDI, KPN, Orange, SoftBank, Bouygues Telecom, Telecom Italia, Rakuten Mobile, Telstra and the vendor Nokia. Separately, three information-assurance agencies (the UK NCSC, Germany's BSI and the US NSA), together with NPL, ask SA2 to endorse that "the 6G architecture will permit core network functions to be isolated from agents, and to operate in their absence" (S2-2611681). Their paper calls embedding agents within critical network functions "an unacceptable level of risk".

Our reading. The compromise settles what both operators and the agencies asked for on availability: connectivity works without AI. It does not settle what the agencies ask for on structure. If the internal organization of AI functionality is left to implementation, and AI "may assist or optimize" connectivity procedures, then the text does not say whether an agent may run inside a connectivity NF. That is the question the agencies care about. CAICT, which on 4 November described the architecture as "basically shaped" (see our Weekly W45), co-signs the agent-based proposal. The documents support a narrower claim: agreement on the floor, not yet on agents, NAS routing or intent transport. The draft reply to the SA plenary (S2-2612402) is still a placeholder: "We assume that there will be some agreement regarding the overall architecture."

The second deadline: voice on 5G, data on 6G

In September the RAN plenary narrowed 5G/6G dual registration to one scenario: two separate single registrations, 5G voice (VoNR) on the 5G leg and 6G data on the 6G leg, network-controlled mobility and no RAN impact. It asked SA2 to determine feasibility by RAN#114 and expects SA to decide no later than SA#114 (RP-262264). SA2 has two opposite draft replies, and the chair's pre-meeting notes label them by outcome:

  • "If dual reg NOT supported", the baseline drafted by Nokia (S2-2612553). It says the phone would have to keep simultaneous connections, sessions and paging on both radios, and that EPS interworking and idle-mode mobility are unresolved. It concludes that SA2 "does not consider it feasible", and that the scenario can be served by a dual-SIM device.
  • "If dual reg supported", drafted by China Mobile (S2-2611816). It says the scenario "shall be supported with UE using single USIM" and is feasible with IP address preservation and no RAN impact. A matching solution comes from China Mobile, China Telecom, vivo, Boost Mobile and EchoStar (S2-2611814).

Also on the 6G agenda

  • Data framework (KI#21). In Prague SA2 agreed that the framework "shall be based on 6G NF(s)". The merged Calgary baseline (S2-2612636) still carries two alternatives. One is a single data-framework NF (Alt A1), proposed by Ericsson, AT&T, Verizon, T-Mobile USA, NTT DOCOMO, Google and others. The other is three standardized NFs for control and discovery, storage, and transfer and processing (Alt B2), backed by vivo, China Mobile, Huawei, ZTE and others.
  • Terrestrial–satellite continuity (KI#23). A merged text led by Ericsson (S2-2612651) covers terrestrial and satellite networks run by different operators. With a roaming or network-sharing agreement, existing mobility procedures apply. Where inter-PLMN handover is not supported, the UE context is shared in advance. With no agreement, the satellite network can be used as an underlay through untrusted non-3GPP access. Multi-orbit access, satellite backhaul and core functions on board satellites are handled in the same agenda item.
  • 6G network for AI (KI#19). Baselines for interim agreements on how AI agents on phones register with the 6G core, find each other and communicate, with agent identifiers and authorization (for example S2-2612618). The consolidated architecture defers their entities and interfaces to this key issue.

What to watch

  • Whether the rapporteurs' clause 8 is agreed as a whole or split. NAS routing and intent transport are the parts most likely to be split off.
  • Whether the agencies' isolation statement is endorsed, noted, or folded into the architecture text.
  • Which dual-registration reply goes out. It decides whether early 6G phones can keep calls on VoNR without a second SIM.
  • Timing. Under the draft session plan, documents handled in parallel sessions can only be "agreed"; approval happens when no session runs in parallel, with a block approval on Friday afternoon (local time). The final chair's notes appear shortly after the close.

RAN1#127 meets in the same week in Calgary. It inherits the radio questions deferred from Jeju, including the length and default period of the 6G synchronization block and whether one transport block may span 400 MHz (see our RAN1#126-bis summary). The outcomes go to the plenaries in Boston: CT#114 on 7–8 December, RAN#114 on 7–10 December and SA#114 on 8–11 December. If SA2 does not converge, its next meeting is an e-meeting on 25–29 January 2027.

Sources

SA2#178 agenda (S2-2611404 r3) and draft session plan, pre-meeting chair's notes (10 November 2026, 19:30), the Calgary invitation and the contributions linked above, read on 12 November 2026 with tracked deletions removed. Several merged baselines still list co-signers as unconfirmed, so this article names only confirmed sources. Plenary dates are from the 3GPP meeting calendar. Positions are what companies propose, not 3GPP decisions.