Introducing AI Cyber Model Arena: A Real-World Benchmark for AI Agents in Cybersecurity

Wiz Research introduces the AI Cyber Model Arena, a benchmark of 257 real-world offensive security challenges across zero-day discovery, CVE detection, API and web security, and cloud security, evaluated via a multi-agent x multi-model matrix with deterministic scoring (rubrics, endpoint/severity matching, lag capture), best-of-three pass@3 runs, and network-isolated Docker environments with standardized tooling; results show offensive capability is jointly determined by model and agent and is highly domain-specific, with ongoing updates planned.