Getting Started
Welcome to the AI Security Bootcamp hands-on lab. You'll build Zero Trust foundations from zero, then protect AI-powered applications and govern workforce AI usage using Cloudflare.
What You'll Build
Over 7 modules you will:
- Set up Zero Trust — configure SAML IdP, reusable Access policies, and CF1 Client / WARP
- Explore a vulnerable AI-powered e-commerce assistant
- Detect attacks using AI Security for Apps
- Mitigate them with WAF rules
- Govern workforce AI usage with Gateway and DLP
- Secure sanctioned MCP access via portal and block shadow MCP
- Operate AI inference with AI Gateway — guardrails, dynamic routing, and cost controls
Lab Components
You'll need the following resources, all provided for you:
| Component | Description |
|---|---|
| Cloudflare account & zone | A dedicated lab account with WAF, Zero Trust, and all required entitlements |
| Windows 11 client | A managed Windows workstation for CF1 Client / WARP enrollment |
| SAML identity provider | A shared SAML-based IdP with preconfigured test users |
| AI Shopping Assistant | A prebuilt e-commerce app with an LLM-powered chat assistant |
| KiwiCart MCP Server | A shared remote MCP server with read-only tools |
| AI Gateway Explorer | A companion app for testing AI Gateway features (M6) |
| Lab guide | This step-by-step guide with validation checkpoints |
Your Lab Slug
Throughout this guide, you will see references to your lab slug. You can find it in the name of your Cloudflare account — it is the two-word suffix after the hyphen (e.g., ... - adjective-noun).
Lab Architecture
The lab uses two control planes across multiple traffic paths:
Inbound (AI App Protection)
You (browser) → Cloudflare WAF + AI Security for Apps → AI Shopping Assistant + LLM
Outbound (Workforce AI Governance)
Win11 client (WARP/Cloudflare One Client) → Cloudflare Gateway + DLP → Public AI Tools
Sanctioned MCP Path
MCP Inspector → MCP Portal → Cloudflare Access (auth) → KiwiCart MCP Server
Unsanctioned MCP Path (blocked)
MCP Inspector → Gateway/SWG → Direct MCP Server URL → BLOCKED
Identity & Trust Model
| Layer | Component | Purpose |
|---|---|---|
| Identity | SAML IdP | Provides user identities for Access and MCP portal auth |
| Access | Reusable policies | Controls who can reach protected resources |
| Client | CF1 Client / WARP | Routes device traffic through Gateway for policy enforcement |
| Gateway | HTTP policies + DLP | Enforces AI usage governance and blocks shadow MCP |
| MCP Portal | AI controls | Provides sanctioned, identity-aware MCP access |
What You Will Build From Zero
In M0, every attendee creates the Zero Trust foundation:
- Configure SAML identity provider — register and test the IdP
- Create reusable Access policies —
All employeesandIT admins - Set up CF1 Client / WARP — enroll the Windows 11 client, connect to Gateway
Do not skip M0. The Gateway AI governance (M4) and MCP portal (M5) labs require a working identity provider, Access policies, and WARP/Cloudflare One Client connection.
Module Overview
| Module | Title | Time | What You'll Do |
|---|---|---|---|
| M0 | Zero Trust Foundation | 35 min | Configure SAML IdP, Access policies, CF1 Client / WARP |
| M1 | Explore & Attack | 20 min | Enable AI Security for Apps, label endpoints, then send normal and adversarial prompts to the AI app |
| M2 | AI Security — Monitor | 30 min | Configure custom topics, inspect detections from M1 traffic |
| M3 | AI Security — Mitigate | 25 min | Apply WAF rules per detection type, compare before/after |
| M4 | Zero Trust AI Governance | 30 min | Gateway AI controls, DLP prompt inspection, Shadow AI |
| M5 | Sanctioned MCP Portal | 40 min | Configure MCP portal, Access policy, block direct MCP |
| M6 | AI Gateway Operations | 50 min | Create AI Gateway, guardrails, DLP, dynamic routing, caching, rate limiting, spend limits |
How to Use This Guide
Each module page includes:
- Task — what you're doing in one sentence
- Why — why this matters
- Steps — numbered steps with exact dashboard paths
- Expected Result — what you should see
- Validation — checkboxes to confirm success
- Troubleshooting — common issues and fixes
Complete each module's Validation step before moving to the next module.
Prerequisites
- A laptop with a modern web browser (Chrome, Firefox, or Edge)
- Your lab credentials (provided by the facilitator)
- Access to the Windows 11 VM
- Node.js 18+ to be installed (for MCP Inspector in M5) in provided Windows 11 VM
Need Help?
- Check the Troubleshooting section in each module
- Raise your hand — facilitators are available to help
- If the dashboard looks different from the screenshots, check that you're in the correct account/zone
Some of the dashboard screenshots may look different from what you see in your browser. This is may due to recent updates to the Cloudflare dashboard. However the functionality and steps should remain the same. Feel free to ask help if you are stuck.