WarRoom - Arena for AI Agent security

88% of enterprises experienced AI agent security incidents in 2026, yet most deploy agents with zero pre-deployment testing. WarRoom is the pre-flight security wargame that changes this. Describe your AI agent, and WarRoom launches a live adversarial battle: a Gemini-powered attacker generates 15 targeted attacks across 5 categories (prompt injection, data exfiltration, privilege escalation, obfuscation, and multi-turn social engineering). A dual-layer defender - Veea Lobster Trap for sub-millisecond regex DPI plus Gemini semantic classification — blocks attacks in real-time. After the battle, WarRoom delivers: a Security Score (0-100), dollar-denominated risk exposure, compliance violation mapping across OWASP Agentic Top 10, NIST AI RMF, EU AI Act, HIPAA, SOC 2, and MITRE ATLAS. Then it auto-generates a hardened Lobster Trap policy that provably blocks the discovered gaps — with before/after proof. What takes security teams weeks of manual threat modeling, WarRoom does in 60 seconds — with only 4 Gemini API calls. The output is a deployable security policy artifact, not just a report. Built with Gemini 2.5 Flash-Lite, Veea Lobster Trap, Next.js, FastAPI, and Docker.