ITU adopts standard for monitoring AI systems in telecom networks
A new ITU recommendation sets out how artificial intelligence systems in telecom networks should be monitored, focusing on performance, reliability, and long-term accuracy.
The International Telecommunication Union (ITU) has adopted a new technical standard, Recommendation ITU-T Q.4081, which defines how AI systems should be monitored in telecom networks. The recommendation was approved in January 2026 and later published on ITU website in March.
The document addresses a practical problem. Many modern telecom networks, including 5G systems, use AI to manage traffic, detect faults, and optimise performance. These systems rely on models trained on past data. However, real-world conditions change over time, which can cause these models to become less accurate.
The recommendation explains that continuous monitoring is needed to ensure that AI systems continue to work correctly after deployment. This means regularly checking how well a model performs, whether the data it receives has changed, and whether its predictions remain reliable.
To address this, the standard defines different types of monitoring. These include checking input data, evaluating model performance, tracking resource use such as computing power, and identifying ‘drift’. Drift refers to situations where the data or environment changes in a way that reduces the accuracy of the model over time.
The document also introduces specific methods and metrics for measuring these aspects. For example, it describes how operators can detect changes in data patterns, monitor prediction accuracy, and identify anomalies that may signal problems in the system.
A framework outlined in the recommendation shows how monitoring systems collect data from networks, analyse it, and feed results back into the system to improve performance. According to the diagram in the document, this process involves data collection, analysis, and result presentation layers working together to support decision-making.
The recommendation is intended as guidance rather than a mandatory rule. It provides a reference for telecom operators and developers on how to maintain reliable AI systems in dynamic network environments, where conditions such as user behaviour, traffic patterns, and infrastructure can change over time.
