AI-Native Core Network Functions in 5G-Advanced (Rel-19) & 6G
(Cloud-Native 5GC, 5G-Advanced, Operations & Capstone)
Why AI-Native Core Functions Matter in 2026
In Rel-19, AI/ML moves from “add-on analytics” (Rel-16/18 NWDAF) to embedded intelligence inside the 5G Core.
In 6G (Rel-20/21 studies), the core becomes AI-Native from Day 1 — not retrofitted, but designed with AI agents, intent-based orchestration, closed-loop automation, and AI-as-a-Service (AIaaS) as first-class citizens.
Key distinction (2026 definition):
- AI-Assisted / AI-for-5G: External models or offline tools (e.g., basic NWDAF analytics).
- AI-Native: AI is embedded in every network function (inference, training, decision-making), with continuous learning, predictive actions, and zero-touch closed loops.
This module equips architects to evaluate vendor implementations, design future-proof 5GC upgrades, and quantify ROI (energy savings, signaling reduction, slice autonomy).
AI-Native Enhancements: Rel-19 Capabilities vs 6G Vision
| Network Function | Traditional / Pre-AI Behavior (5G Rel-16/18) | Rel-19 AI-Native Enhancements (Production-Ready 2026) | 6G AI-Native Vision (Rel-20/21 Studies) | Business / Operational Impact |
|---|---|---|---|---|
| NWDAF (Network Data Analytics Function) | Analytics add-on; limited to descriptive/predictive reports | Phase 2: Full model training/inference orchestration, Federated Learning, Model Transfer, abnormal/excessive signaling prediction, policy assistance to PCF, accuracy reporting, vertical federated learning with AFs | AI orchestration hub with embedded agents; distributed inference/training; AIaaS exposure | Closed-loop automation; 30–40% signaling reduction; predictive maintenance |
| PCF (Policy Control Function) | Static or rule-based policies; manual QoS/URSP updates | AI/ML-driven dynamic policy & QoS (uses NWDAF predictions for QoE, QoS parameter sets); energy-aware & slice-load policies | Intent-based policy orchestration; AI agents translate business intent → real-time policies | Zero-touch slice management; predictive QoE for XR/URLLC; energy KPIs as SLA |
| AMF (Access & Mobility Management) | Reactive registration/mobility handling | Predictive signaling control; signaling storm detection/mitigation via NWDAF | AI agents for proactive mobility & anomaly handling; intent-driven registration | Massive IoT scale; reduced signaling storms; self-healing mobility |
| SMF (Session Management) | Rule-based PDU session & QoS flow control | AI-assisted session optimization; dynamic edge steering with predictive QoS | Native compute steering + AI-driven session lifecycle | Lower latency for industrial/edge use cases; autonomous PDU sessions |
| UPF (User Plane Function) | Packet forwarding & basic QoS enforcement | AI-informed traffic steering & optimization (via NWDAF/PCF); protocol for AI data collection from UPF | Integrated sensing + compute in UPF; AI-native packet processing | ISAC-enabled services; real-time AI workload offload |
| Other NFs (UDM, NRF, AUSF, NSSF) | Static discovery & management | Enhanced data exposure for AI models; roaming analytics support | Full AI-native SBA with intent layer + modular NAS | Simplified multi-vendor & cloud-native operations |
Sources (April 2026): 3GPP TR 23.288 (NWDAF), TS 23.288 Rel-19 updates, TR 23.801 (6G Core study), Ericsson/Nokia/Qualcomm MWC 2026 white papers.
Key Architectural Changes in AI-Native 5GC
- Data Collection & Exposure — DCCF/MFAF/ADRF + new AI-specific interfaces; continuous telemetry from every NF/UE.
- Model Lifecycle Management — Training (MTLF), inference (AnLF), federation, transfer, and accuracy monitoring become standardized services.
- Closed-Loop Automation — NWDAF → PCF/AMF/SMF feedback loops for self-optimization (energy, load, QoS).
- Intent-Based Networking — 6G introduces an Intent Layer where operators state “what” (e.g., “maximize energy efficiency for IoT slice”) and AI agents handle “how”.
- Federated & Distributed AI — Models train across NWDAFs without sharing raw data (privacy + scale).
Key Takeaways for AI Architects & Operators
- Rel-19 makes AI-Native practical today (focus on NWDAF + PCF integration).
- 6G makes it foundational (AI agents + intent layer in every NF).
- Biggest ROI areas: signaling storm prevention, predictive QoS/energy policies, autonomous slicing.
- Cloud-native 5GC deployed in 2026 is the perfect foundation — add AI model transfer and federated learning now for seamless 6G evolution.
Key Insights (from sources)
- 5G-Advanced (Rel-18/19) introduces AI/ML deeply into core and RAN for automation, optimization, and analytics
- Rel-19 expands AI use cases such as network slicing, coverage & capacity optimization, and positioning
- AI-native architecture enables self-optimizing, autonomous, and intelligent networks, forming the foundation for 6G
- 6G will be AI-native by design, embedding intelligence across all network layers and lifecycle