We Let an AI Agent Run Autonomous Experiments on ModSecurity CRS. Here's What It Found.
Starting from stock OWASP CRS defaults, an AI agent improved balanced accuracy from 86.7% to 96.7% on CRS v3.3.8, and from 80.8% to 98.4% on CRS v4.24.0, running 30 experiments over 7 hours. Full methodology, results, and downloadable configs.
