ITU launches telco agent challenge

The International Telecommunication Union has opened a new contest for artificial intelligence agents that can diagnose and fix telecom network faults. (aiforgood.itu.int) The challenge is run with the Global System for Mobile Communications Association, the European Telecommunications Standards Institute, the Institute of Electrical and Electronics Engineers Emerging Technologies Initiative GenAINet, and TM Forum. The official launch date was March 3, 2026, followed by a global webinar on March 20. (aiforgood.itu.int) Telecom troubleshooting is the work of finding the cause of an outage or service problem across routers, software, and network links. The International Telecommunication Union says modern networks are now so software-driven that failures can cascade across layers and services faster than manual operations teams can respond. (aiforgood.itu.int) The new contest asks teams to build “agentic” systems, meaning software that does not just answer a question but can use tools, take multiple steps, and act on changing network conditions. The benchmark focuses on Internet Protocol networks and three tasks: topology restoration, path query, and fault localization. (aiforgood.itu.int) The International Telecommunication Union says entries will be tested in a multi-vendor “Agent Sandbox,” a simulated environment built to mimic real telecom operations. A central orchestration layer will feed structured questions to each system, connect it to domain-specific application programming interfaces, and score accuracy, efficiency, and robustness. (aiforgood.itu.int) This contest follows an earlier telecom language-model competition announced on November 25, 2025. That earlier challenge focused on fine-tuning large language models for root-cause analysis using the TeleLogs dataset, which the organizers said contains about 4,000 multiple-choice troubleshooting questions. (gsma.com; aiforgood.itu.int) The earlier competition offered three tracks: generalization to new faults, small models at the edge, and explainability and reasoning. The new 2026 contest shifts from training models on a fixed question set to testing agents that can reason over network state and call tools in real time. (gsma.com; aiforgood.itu.int) Prize money has increased as well. Zindi lists the new Telco Troubleshooting Agentic Challenge at €40,000, while the earlier AI Telco Troubleshooting Challenge was listed at €35,000. (zindi.africa; gsma.com) The organizers say winners will present their systems at Mobile World Congress Shanghai, scheduled for June 24 to 26, 2026. The competition itself runs from March through June. (aiforgood.itu.int; aiforgood.itu.int) The pitch is straightforward: if these agents can localize faults faster and restore network paths with less human intervention, carriers cut downtime and operating costs. The International Telecommunication Union is now trying to turn that idea into a benchmark other telecom groups can test against. (aiforgood.itu.int; aiforgood.itu.int)