Telco groups open challenge to test LLMs on real network faults

A new industry challenge is calling on telecom operators, AI researchers, and startups to build LLMs that can identify and explain the root causes of network faults – a persistent problem that costs operators millions each year. The AI Telco Troubleshooting Challenge launched today with support from GSMA, ETSI, IEEE GenAINet, ITU, and TM Forum, bringing together a wide mix of organisations that want to push AI deeper into network operations.

The challenge comprises of several tracks, each focused on a different aspect of network troubleshooting. One will test how well an LLM can handle unfamiliar or previously unseen faults. Another looks at small, efficient models that can run at the network edge, where compute resources are more limited. A third evaluates how clearly an AI system can show its reasoning when diagnosing issues. Additional categories will cover edge-cloud security and the development of AI services for application builders.

Organisers see the challenge as a way to surface practical approaches that help operators restore service faster and automate more of their workflows. Entrants will be judged on accuracy, speed, reasoning ability, and security. Alongside the main community partners, the challenge is supported by Huawei, InterDigital, NextGCloud, RelationalAI, xFlowResearch, and technical advisers from AT&T.

Industry groups outline their expectations for AI in network operations

ETSI says many of the problems facing operators today come down to long-standing gaps in model generalisation and the constraints of edge computing. Dario Sabella, Chair of ETSI MEC, said the challenge gives teams access to specialised datasets and infrastructure that can “accelerate the adoption of Telco AI.” He highlighted that small, efficient language models at the edge may help make AI deployments more accessible in the industry.

IEEE GenAINet views the challenge as an opportunity to test ideas central to building more autonomous networks. Prof. Merouane Debbah, General Chair of IEEE GenAINet ETI, said it focuses attention on issues like unseen fault detection, interpretability, and edge-efficient AI – areas he said are essential for “making AI-native telecom infrastructures a reality.”

The ITU sees this challenge as part of its work to lower the barriers for innovators who want to experiment with telecom-focused AI systems. Seizo Onoe, Director of the ITU Telecommunication Standardisation Bureau, said the organisation’s global challenges aim to give developers access to the computing resources, datasets, and mentors needed to help their work reach “meaningful impact.”

The new initiative builds on previous work to benchmark AI models for telecom tasks. That includes curated datasets like TeleLogs and tools developed under the GSMA Open-Telco LLM Benchmarks community, which also maintains a public leader board showing how different models perform on telco-specific scenarios. The benchmarks have given operators and researchers a clearer view of how well LLMs handle tasks like fault diagnosis, trend analysis, and reasoning under uncertainty.

GSMA says the industry’s interest in better reasoning and diagnosis stems from the scale of financial losses caused by network faults each year. Louis Powell, Director of AI Technologies at GSMA, said root cause analysis “is an important pain point for operators,” and improving it could both strengthen reliability and reduce ongoing operational costs.

He said the challenge aims to push the development of models that combine reasoning ability with efficiency and scalability.

AT&T, one of the challenge’s technical advisers, recently ran its own experiments that point to what smaller AI models are capable of. The company fine-tuned a 4-billion-parameter model that outperformed all other models on the TeleLogs RCA benchmark, including frontier systems like GPT-5, Claude Sonnet 4.5, and Grok-4. Andy Markus, Chief Data Officer at AT&T, said the challenge brings together “an important business problem and a technical opportunity,” and that broader collaboration in the industry can help take the work further.

The organisers say the challenge offers a space for teams to test new ideas with access to resources that are often hard to secure independently. As network complexity increases, they expect more operators to explore AI systems that can diagnose issues fast, explain their reasoning, and run efficiently in edge and cloud environments.

For many of the partners involved, this challenge is a step toward building telecom networks that are more reliable, more adaptive, and easier to manage at scale.

(Photo by Google DeepMind)

See also: How fragmented regulation stifles mobile security innovation

Want to discover how IoT is transforming telecoms and connectivity? Join the IoT Tech Expo in Amsterdam, California, and London. Explore how innovations in 5G, edge computing, and IoT are shaping the future of networks and services. The event is part of TechEx and co-located with other leading technology conferences, click here for more information.

Telecoms News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.

Tags: ai, at&t, edge computing, networks