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

The transition to AI-Native 6G architecture, driven by Generative AI (GenAI) and Large Telecom Models (LTMs) embed AI directly into network control loops, unlocking unprecedented automation and autonomous service orchestration. This structural shift introduces critical risks related to data sovereignty, security, explainability, and regulatory compliance when AI governance is external or opaque. This paper introduces organisational Sovereign AI (SovAI) as the strategic imperative for 6G, defining it as the operator’s mandate for end-to-end control over the entire AI lifecycle within its domain. We propose a four-pillar architectural and operational framework for realising SovAI, focusing on the O-RAN architecture. The framework details how SovAI is enforced by architecting for Trust, mandating verifiable control points within the Near-RT and Non-RT RICs to secure xApps/rApps. We introduce the concept of Trusted Telecom Copilots for operational governance and detail how Federated Learning integrated with Privacy-Enhancing Technologies (PETs) ensures secure R&D autonomy and data sovereignty. Our findings highlight that SovAI is not merely a regulatory concept but a technical prerequisite for trustworthy and resilient AI-Native 6G networks.