Artificial Analysis launches Cyber Index for enterprise AI defense; Grok 4.7 and MiMo-V2.6-Pro reportedly lead

Artificial Analysis has announced the Artificial Analysis Cyber Index and the Artificial Analysis Cyber Index Alliance as what it describes as a new standard for evaluating AI models on enterprise cyber defense, with Collinear AI, IBM, NVIDIA, and Vercel named as launch partners. Artificial Analysis says the index combines three benchmarks — CWE-Bench-AA, DeepsecBench-AA, and CyberGym-E2E-AA — to measure how well agents find and fix vulnerabilities. In results shared by the company, Grok 4.7 (xhigh) and MiMo-V2.6-Pro tied for the top score at 56. The article also says the best model identified 41% of expert-verified issues in DeepsecBench-AA, and that safety refusals limited several frontier models, including full refusals on every CyberGym-E2E-AA task from GPT-6 Sol and GPT-6 Astra.