Page 82 - AI Standards for Global Impact: From Governance to Action

AI Standards for Global Impact: From Governance to Action

11�6 Opportunities for networking standards

The AI and Machine Learning in Communications Workshop highlighted several potential
opportunities for the standardization work of ITU-T Study Group 13 (Future networks), especially
given ongoing advances in AI, ML, and the shift towards IMT-2030 (6G) networks.

1) AI-native network architectures
o There is a strong need for standardized frameworks for AI-native network architectures,
including scalable models for intelligent RAN, core, and edge networks.
o Future standards could focus on embedding AI agents and pipelines across all network
layers, transitioning from “AI add-on” approaches to fully integrated, intelligent, and
adaptive communication systems.
o Workstreams such as intent-driven automation (multi-agent systems) and digital twin
architectures present further potential avenues for standardization and benchmarking.

2) Datasets, benchmarking, and validation
o Standardization of datasets and benchmarks for AI/ML in networks is becoming
increasingly critical, with the need for trusted, multimodal, and open datasets for
training, validation, and certification of AI models.

3) Open-source and collaborative ecosystems
o Open-source toolkits, platforms, and simulation environments could support rapid
prototyping and interoperability between standards bodies and industry partners.
o Open-source projects could accelerate feedback loops from real-world deployments
into the standardization process, helping ensure specifications stay relevant.

4) Model, agent, and API standardization
o Frameworks for agent-centric architectures (e.g., agent-first control planes and APIs
for agent communication in network functions) are emerging as a hallmark of next-
generation networks and a clear potential area for new standards.
o Standardizing interfaces such as A2A communications, Model Context Protocols,
and federated learning frameworks could support the interoperability and lifecycle
management of AI functionalities in telecom.

5) Generative AI, LLMs, and domain-specific AI
o Benchmarking frameworks for generative AI in telecom are needed and could focus
on requirements, deployment methodologies, and domain-specific adaptations (e.g.
telecom LLMs and RF-centric models).

6) IMT-2030 (6G) Standardization
o Future SG13 standards could support IMT-2030 by defining requirements and
functional architectures for network resource sharing, autonomous networks, and
advanced fixed-mobile-satellite convergence scenarios.
o Cross-layer knowledge integration, modular AI pipelines, collaborative inference, and
secure model distribution could be key areas for standards work.

70