Evolving to the AI-Native 6G Core: Why Architecture Decisions Cannot Wait
The journey to 6G is no longer a distant research exercise. The timetable is compressing, and the industry has a narrow window in which to turn promising concepts into mature standards, interoperable products and deployable systems. Before 2029, standards freezes, spectrum decisions, commercial silicon, devices, network architecture, security frameworks and credible use cases all need to converge.
That is an aggressive sequence, particularly because these dependencies cannot progress in isolation. A delay in spectrum, silicon, architecture or governance can quickly affect everything downstream. It is why the decisions being made now about the 6G system architecture and core matter so much, as they will determine whether the first 6G services arrive with a strong foundation or inherit compromises that could take years to resolve.
Learning from the 5G transition
The first question is a fair one: why does 6G need a dedicated core at all?
The answer begins with experience. One of 5G's most consequential structural choices was to launch initially in non-standalone mode, anchored to 4G, and migrate to standalone later. That approach accelerated early radio deployments, but it also delayed access to many of 5G's architectural benefits. The industry does not want to repeat that pattern. The working ambition for 6G is standalone from day one, supported by a dedicated 6G core.
There is also a functional reason. 6G is expected to make capabilities such as AI-native control, integrated sensing and communications, tightly integrated compute and native non-terrestrial networking part of the architecture itself. At the same time, it must address limitations already visible in the 5G core, including signaling and service-based architecture scaling, evolving security needs, inefficient resource handling and QoS assumptions that do not fit emerging AI traffic.
This does not imply a clean-sheet redesign. The pragmatic path favored is to reuse and evolve proven 5G core network functions where that makes sense, introduce new functions only where they are justified, and enhance the service-based architecture rather than replace it wholesale. The objective is evolutionary continuity on a clean standalone foundation.
5G core vs 6G core architecture
A core that brokers, authorizes and orchestrates
Three use-case families illustrate how the core's role will expand.
For integrated sensing and communications (ISAC), the 6G core may discover and select sensing functions, authorize sensing services, transport and expose sensing data, and coordinate mobile sensing entities.
For ubiquitous connectivity (UC), it must maintain service continuity across terrestrial and non-terrestrial access, interwork with existing systems and apply policy consistently across domains.
For AI and communications (AIAC), it may support discovery and authorization between AI agents, expose network capabilities through intent-based interfaces, provide inferencing or training services, and adapt policy, QoS and compute selection to AI workloads.
These are not separate worlds. A robotics service, for example, could simultaneously depend on sensing, seamless mobility and coordination between AI agents. The 6G core therefore becomes an active broker, authorizer and orchestrator across data, compute, connectivity and intelligence. That is a broader responsibility than the 5G core was designed to carry.
Architecture is taking shape, but remains open
The architecture is still under study. 3GPP work is distributed across several groups, with SA2 acting as the architectural anchor. Its foundational study (TR 23.801-01) has identified multiple key issues and a large field of proposed solutions that now need evaluation before normative work can mature.
An evolved 5G service-based architecture is the likely starting point, but important questions remain unresolved. Should 5G and 6G operate as distinct cores linked through new interworking interfaces, or should they share selected functions? A practical outcome may combine both approaches: common functions where continuity matters, with separate functions interworking where isolation or independent evolution is more valuable.
Candidate changes show the direction of travel. The user plane could converge with edge compute. Mobility and session-management functions could become intent-aware and use AI to make more autonomous decisions. New functions may be required to authorize and orchestrate sensing, manage AI agents, broker agent discovery and communication, and control the collection and processing of data at scale. These remain proposals, not a settled function list, which is precisely why evidence from implementation and testing is needed now.
AI changes both the network and its traffic
AI affects the 6G core in two complementary ways. “AI for 6G” uses AI inside the network for intent processing, dynamic procedure composition, closed-loop learning and progressively more autonomous operations. “6G for AI” turns the network into a platform for external AI agents, providing connectivity, discovery, communication, capability exposure and potentially AI services.
Standards discussions are considering three broad architectural patterns: embedding AI in existing network functions, introducing dedicated AI network functions, or creating a distinct AI domain alongside the traditional core. Each presents trade-offs in latency, signaling, governance, interoperability, architectural complexity and speed of evolution. Given the timetable, a hybrid model may prove attractive, but it must be validated rather than assumed.
Meanwhile, AI is transforming what networks carry. Traditional traffic is comparatively predictable, often downlink-heavy and shaped by time-of-day demand. AI traffic can be bursty, uplink-heavy, latency-sensitive and persistent. Applications are becoming AI workloads; networks are becoming interconnects among agents, models, edge inference and AI factories; and service metrics must evolve accordingly. Alongside throughput, latency and availability, operators may need to understand time to first token, energy per token, token efficiency and compute-to-action time.
That shift also challenges static QoS. 6G will need dynamic adaptation during a session, graceful degradation when resources change, richer monitoring and bidirectional negotiation between the network and application. The network can no longer set policy once and expect an AI-driven application simply to live with it.
Trust must be engineered continuously
An AI-native core will only succeed if operators can trust its decisions. Governance cannot be limited to model approval before deployment. Networks will need runtime guardrails, evidence, accountability and mechanisms that allow autonomy to increase only as confidence grows.
A continuous assurance loop can combine digital twins, active probing and passive telemetry. A digital twin can emulate a proposed AI-driven policy before it reaches live traffic. Active probing can validate behavior under controlled conditions. Passive telemetry can then observe real decisions and outcomes, building the traceability record needed for governance. If performance drifts or a guardrail is breached, the system can reduce autonomy, fall back or re-plan, then repeat the validation cycle before the change is trusted again.
Prototype now to make better standards decisions
With the standards window narrowing, the industry's priority should be rapid prototyping of the hardest architectural questions. At Keysight, our Core Network Testbed Platform is designed to support that experimentation through realistic, at-scale emulation and high-capacity traffic generation, including AI traffic. It can help researchers, vendors, operators and standards bodies evaluate new network functions, service-based architecture behavior, security, QoS and end-to-end performance for sensing and non-terrestrial services.
Working prototypes turn theoretical trade-offs into measurable evidence. They expose scaling limits, interoperability risks and operational consequences while designs are still changeable. That evidence can help the industry protect capabilities that will make 6G valuable on day one and avoid well-intended architectural decisions that become expensive constraints later.
The 6G core should evolve pragmatically, but ambition must not be confused with complexity. The goal is a standalone, AI-native foundation that reuses what works, adds what is genuinely needed and earns trust through continuous assurance. The timetable demands that we learn earlier, together.
For a deeper dive into this topic, watch our webinar on demand.