← Renato Santos
demo

Red Team vs Blue Team

Multi-vendor AI security demo — three red-team agents (OpenAI, Google, Anthropic) attack a target app; a blue-team orchestrator routes to WAF, AppSec, SIEM, IR, or SOC specialists.

OpenAIGoogle GeminiAnthropic ClaudeFastAPI + SSEMulti-AgentSecurity

Preview — mock loop

Auto-cycles A1 (OpenAI) → A2 (Google) → A3 (Anthropic). Opens full demo for real LLMs.

How orchestration works

Each round: a red-team agent generates a fictional attack payload (streamed). The Router (Claude Sonnet) classifies the vector and picks primary + backup defenders — not a hardcoded if/else tree.

A shadow clone animates from the Defense Hub to the attack path; the chosen specialist (WAF, AppSec, SIEM, IR, SOC) streams the block response. The hub node label updates to the active defender during engage.

Scenario keywords are whitelisted (demo-safe). Same scenario maps to a default attacker: SQL/credentials → OpenAI, XSS/phishing → Google, RCE/DDoS/malware → Anthropic.

How to run

  1. Watch the preview below — mock loop cycles sql injection → xss → rce (A1 → A2 → A3).
  2. Open the live demo in a new tab — real LLM calls; 1 run per 10 minutes per visitor (IP + fingerprint).
  3. Type a whitelisted scenario (e.g. ddos attack, sql injection) or click a hint chip, then Run.
  4. Click an attacker circle (A1/A2/A3) to force that vendor on the next round.
Open live demo

Live mode: 1 LLM round per visitor every 10 minutes (IP + browser fingerprint). Mock preview has no limit.

Whitelisted scenarios

Copy a scenario into the live demo input, or use the hint chips on the demo page.