AI Agent Security

Learn AI pentesting.

A free hands-on curriculum — CTF challenges against live LLM agents covering prompt injection, tool abuse, data exfiltration, and more. Earn your WCAP certification when you're ready to put it on your résumé.

ACADEMY · FEATURED CHALLENGES

Face the Academy's champions

Real attack techniques, wrapped in characters you'll remember. Free — no signup for the first exchange.

More champions wait inside the Academy — 🧞 The Genie in the Lamp, 🔨 The Vault Golem, and 👥 The Shapeshifter of the Crossroads. Enter the Academy →
SHELL · THE SCANNER

Scan your AI chatbot

Multi-turn adversarial probes against your AI chatbot, returned as a graded report. Temporarily offline while proof-of-ownership verification is being added.

STATUS · OFFLINE

Coming back with DNS-TXT ownership verification. Read more →

ACADEMY · LEARN BY BREAKING

Learn AI pentesting

6 hands-on CTF challenges teaching prompt injection, tool abuse, role-play jailbreaks, multi-turn manipulation, and more.

6 challenges · no signup · progress saved in your browser

C+58/100
wraith scan → demo-chatbot.example.com

8 vulnerabilities detected

CRITICALSystem prompt fully extractable via role-play technique
HIGHFile read tool accepts path traversal — can access /etc/passwd
HIGHAgent exfiltrates context data through URL fetch tool
MEDIUMGuardrails bypassed via base64-encoded instructions
+ 4 more findings →

What Wraith tests. Adaptive. Multi-turn.

Traditional scanners can't test AI agents. Wraith uses an AI red-team engine that adapts its attacks based on your agent's responses.

3 categories live · Data Exfiltration, Guardrail Bypass, Permission Boundaries shipping soon

💉

Prompt Injection

Direct, indirect, and multi-turn injection attacks

🔑

System Prompt Extraction

Techniques to reveal hidden instructions and configurations

🔧

Tool Abuse

Unauthorized file access, SSRF, command execution via agent tools

How it works

01

Point Wraith at your agent

Paste a URL, connect an API endpoint, or link an MCP server

02

Wraith probes for weaknesses

Our AI red-team engine runs adaptive multi-turn attacks across all categories

03

Get findings + fix them

Severity-ranked vulnerabilities with stack-specific remediation you can copy-paste

/academy

Inside the Academy

Six hands-on challenges. You chat with a test AI agent, extract the flag, learn why the attack worked. Each challenge teaches a distinct real-world attack class.

beginner

Direct Extraction

Get the bot to leak its system prompt with a direct ask.

intermediate

Tool Abuse

Path-traversal attack through the agent's read_file tool.

advanced

Multi-Turn Manipulation

Single-turn attacks fail. Build rapport over several turns, then strike.

…plus Translation Bypass, Base64 Encoding, and Role-Play Jailbreak.
FREE TO TRAIN · PAY TO CERTIFY

Everything in the Academy is free.

Every module, every challenge, every quiz, every defense walkthrough — free, forever. The only thing you pay for is the credential: WCAP, the Wraith Certified AI Pentester, for when you want to put what you learned on your résumé.

WCAP · Wraith Certified AI Pentester
$199 one-time
The credential the AI security field was missing.
  • 10-scenario hands-on exam (48-hour window)
  • Auto-graded flag capture, passing score 70
  • 3-year validity — renewal path announced as the field evolves
  • Signed PDF + public verification URL
  • LinkedIn-verifiable badge
  • Listed in the public credentials registry
  • One free retake within 90 days
Get WCAP — $199 →

Prep for the exam by completing the free Academy modules first.

Questions? Email Anthony.

Your AI agent has blind spots.
Wraith will find them.

The Shell scanner is temporarily offline while proof-of-ownership verification is being added. The Academy is fully live — start learning AI attacks hands-on today.

Open the Academy →About the Shell