Grey Noise AI Cyber Index: Benchmarking AI Models on Defensive Capabilities
Summary: GreyNoise launched the AI Cyber Index, which tests 46 closed and open-weight models across three defensive tasks - Network Detections, File Detections, and Attack Forensics - using 145 problems drawn from eight years of labeled traffic and artifacts from its Global Observation Grid. Scores are a weighted composite (Network Detection 55%, File Detection 30%, Attack Forensics 15%), with each model attempting every problem at least three times and refusals scored as zero, and a model counts as capable at 90% of the top score, a 64.5% threshold with refusals penalized.
Why it matters: GreyNoise notes the window between vulnerability disclosure and exploitation has narrowed from days to minutes, and most public cyber benchmarks still measure offensive tasks like exploit development rather than whether a model can write a detection or investigate an attack. With refusals scored at zero, a model that declines to work on real malware earns nothing - so the benchmark measures reliability under the conditions defenders actually face.
Read the original at GreyNoise Blog ↗
- Topics
- Vulnerabilities & Patching
- Related
- Attackers Exploit UnrealIRCd Flaw That Enables Server-Level Command Injection Cyber Risks · Nov 14, 2026U.S. CISA adds Check Point and UnrealIRCd flaws to its Known Exploited Vulnerabilities catalog Cyber Risks · Nov 14, 2026PostgreSQL Patches 52 Flaws, Including 22-Year-Old Bug Enabling Authenticated RCE Cyber Risks · Nov 13, 2026