NIKA: an open network arena for AI agents in network troubleshooting, presented at IEEE LCN 2026

On 9 October Carlo Centofanti gave an invited talk at IEEE LCN 2026 in Coimbra, in the LLMNet special track, during the 50th-anniversary edition of the conference. The title: Building an Open Network Arena for AI Agents in Network Troubleshooting.

AI agents are entering network operations, but they cannot be validated on production incidents: these are rare, impossible to reproduce and too risky to replay on live networks. What is needed is a reproducible, fair and open evaluation, and this is the idea behind NIKA: faults injected into emulated networks, and agents scored with rule-based metrics rather than anecdotes.

NIKA 0.2 is out and the public leaderboard is online: 169 cases, 29 scenarios, 75 fault types. Bring your agent and see where it lands.

• Code: github.com/sands-lab/nika
• Leaderboard: sands-lab.github.io/nika
• SANDS Lab @ KAUST: sands.kaust.edu.sa

NIKA is a joint effort with Zhihao Wang and Dingde Jiang (UESTC), Alessandro Cornacchia and Marco Canini (KAUST), Alessio Sacco and Franco Galante (Politecnico di Torino), and the new members Mariano Scazzariello, Tommaso Caiazzi and the Kathará team. Thanks to the LLMNet track organizers, Suranga Seneviratne and Madhusanka Liyanage, and to the track chairs for the invitation.