Hanif Kukkalli▌

We introduce the AI-Native Control Entry Point (AICEP), a standardless control abstraction that replaces schema-defined management interfaces with learned semantic translation. Instead of enforcing shared control models (e.g., O1/A1/O2) or synthesising APIs on demand, AICEP provides a Single Entry Point API per Network Function (NF) that accepts free-form intents and delegates translation to an external, NF-specific Generative AI as a Service (GAIaaS). The GAIaaS grounds vendor documentation, synthesises version-aware plans, invokes native APIs, and verifies post-conditions in a closed loop.

We generalise this building block into a three-layer AI-native management architecture spanning IBN-enabled SMO, cloud/network controllers, and heterogeneous NFs (RAN/Core/Backhaul), achieving interoperability without introducing new control schemas. A prototype implementation demonstrates feasibility: GPT-5 translates intents into AWS EC2 actions with synthesised user data, achieving deterministic web-server deployment in less than 60.0 s under fixed decoding.

By shifting interoperability from standards to semantic assurance, AICEP reduces onboarding effort, preserves vendor autonomy, and provides a deployable pathway toward AI-native orchestration with guardrails, model routing via Small Language Models (SLMs), and governance centred on verification rather than interface standardisation.

Traditional rule-based or static management approaches struggle to cope with the dynamic, multi-layered nature of 5G/6G networks, creating a strong motivation for AI-native solutions—management systems built from the ground up with artificial intelligence—to enable autonomous, real-time network control. In this work, we introduce GNN-ATIVE, an AI-native orchestration framework that leverages Graph Neural Networks (GNNs) and knowledge graphs (KGs) in a unified graph-based paradigm for network management. GNN-ATIVE uses a semantic knowledge graph to represent the network's state and context, employing standard ontologies to ensure consistency and interoperability. Building on this foundation, we design Knowledge Graph enabled Generative Pretrained Transformer (KG-GPT), a novel graph-to-graph Transformer model that performs knowledge-driven reasoning on the KG. KG-GPT ingests the structured network state (nodes, links, and attributes) and infers optimal configurations or management actions, serving as a high-level decision engine for the orchestrator. We implement and evaluate GNN-ATIVE on an Optical Transport Network (OTN) testbed using real network components. The results demonstrate that GNN-ATIVE can effectively manage OTN resources and adapt to network changes while achieving low-latency inference for decision making.

This paper presents a framework that integrates Generative AI as a Service (GAIaaS) into Service Management and Orchestration (SMO) systems to enable intent-based automation. By leveraging ChatGPT-4o's advanced natural language capabilities, the system interprets user intents and generates policy-driven service chain configurations. Prompt engineering techniques are employed to evaluate the model's performance across key areas, including response time for intent processing, token usage efficiency, repeatability of outputs, infrastructure cost, and multilingual support. The framework consists of various components such as orchestration engine, cloud network function controller, and SDN controller and automates service chain design and resource management. The obtained results demonstrate its reliable and scalable performance across different scenarios. However, challenges related to handling large prompts and sustaining performance under high loads have been identified. This work highlights the potential of GAIaaS to provide scalable, adaptive, and intelligent network automation.

The evolving mobile network specifications, as well as the increasingly stringent requirements for bandwidth, quality of service, and dynamic on-demand adaptability, push the boundaries of what is achievable with legacy networking technologies. Software-defined networking (SDN) and Network Functions Virtualization (NFV) provide agile, cost-efficient solutions by enabling an automated Service Management and Orchestration (SMO), allowing for dynamic resource allocation and network (re-)configuration. In this paper, we present an experimental demonstration of an SMO solution for virtualized mobile networks based on a seamless integration of an SDN-based transport network and an cloud computing environment. Our proposed SMO solution is evaluated by investigating the time required for establishing a virtual mobile core network service chain (i.e. a core network slice) on demand within an SDN-based cloud network considering its specific traffic and QoS requirements.

Dynamic network slicing involving multi-operator network resources is challenging due to lack of cooperation and inter-operability. Currently available slicing techniques focusing on static network configurations without run-time modifications and prioritization are insufficient to provide on-demand, instantaneous network slicing. With currently developed open-source platforms for virtualization and orchestration such as ONAP and ORAN, dynamic network slicing can be enabled. In this paper, a novel dynamic resource allocation scheme for dynamic 5G network slicing (DYSOLVE) is proposed and evaluated for a vehicular emergency scenario. DYSOLVE allocates both the radio as well as transport network resources of multiple network operators cooperatively and aims for slice cost optimization while ensuring service availability and QoS. Our performance evaluation shows significant improvements in the proposed DYSOLVE scheme compared to a traditional baseline approach with fixed resource allocation and without multi-operator cooperation.

Network disaggregation is gaining momentum in the industry for enabling flexibility and choice for network operators. A key challenge is operations and management in multi-vendor environments. SDN architectures are becoming popular for this task. A key prerequisite are well defined and standardized interfaces between network elements/layer and multi-vendor/domain SDN controllers/orchestrators. This paper describes recent progress in ONF core information model definition, its implementation, network architecture integration and related multi-vendor interoperability tests.