Artificial Intelligence is expected to play a central role in future 5G-Advanced and 6G systems. Its applications range from energy optimisation and transceiver simplification to enhanced adaptability, automation, and sensing-enabled services such as JCAS/ISAC. However, mobile networks are critical infrastructures, where reliability, robustness, and trustworthiness must remain extremely high. Deploying AI-based functions directly into live networks without realistic validation may therefore introduce significant operational risks.

At the same time, AI-enabled networks will generate, process, and act upon very large volumes of data, especially at the edge. If not properly managed, this data burden may become comparable to, or even exceed, the user-plane traffic itself.

Digital Twins are emerging as a key approach to address these challenges in 5G and 6G network development. They can provide realistic, flexible, and scalable environments for testing AI algorithms before deployment.

They can also support continuous validation by comparing expected and actual network behaviour during operation.

In this talk, we will discuss how Digital Twins can be applied in the telecommunications domain, from pre-launch validation to performance monitoring and optimisation.
We will present examples covering different network layers, use cases, and levels of abstraction.

The talk will also highlight the limitations of Digital Twin approaches, including modelling accuracy, scalability, data availability, and real-time constraints.
Particular attention will be given to the need for multiple levels of Digital Twins, adapted to different problem scales and validation objectives.

The examples and results presented are based on investigations carried out with network operators and equipment vendors in several European-funded research projects.

The talk will conclude with perspectives on how Digital Twins may become a cornerstone for trustworthy AI deployment in future 5G/6G networks.