GreyNoise launches AI Cyber Index for 46 models

GreyNoise has launched the AI Cyber Index, benchmarking 46 closed and open-weight AI models on defensive security capabilities across network detections, file detections, and attack forensics. Testing 145 problems based on real-world Global Observation Grid data, the index evaluates performance, refusal rates, time, and cost. Results show that models struggle with defensive tasks, and top-performing models like Claude Opus 5.5 and Grok 4.5 face trade-offs between capability and high refusal rates for malware-related inputs. Open article →