Sovereign AI for 6G: Towards the Future of AI-Native Networks

Abstract

The advent of Generative Artificial Intelligence (GenAI), Large Language Models (LLMs), and Large Telecom Models (LTM) significantly reshapes mobile networks, especially as the telecom industry transitions from 5G's cloud-centric to AI-native 6G architectures. This transition unlocks unprecedented capabilities in real-time automation, semantic networking, and autonomous service orchestration. However, it introduces critical risks related to data sovereignty, security, explainability, and regulatory compliance especially when AI models are trained, deployed, or governed externally. This paper introduces the concept of `Sovereign AI' as a strategic imperative for 6G, proposing architectural, operational, and governance frameworks that enable national or operator-level control over AI development, deployment, and life-cycle management. Focusing on O-RAN architecture, we explore how sovereign AI-based xApps and rApps can be deployed Near-RT and Non-RT RICs to ensure policy-aligned control, secure model updates, and federated learning across trusted infrastructure. We analyse global strategies, technical enablers, and challenges across safety, talent, and model governance. Our findings underscore that Sovereign AI is not just a regulatory necessity but a foundational pillar for secure, resilient, and ethically-aligned 6G networks.