🎃
🕷️
👻
🦇
🎃Cybersecurity Awareness Month 2025🔐

October Agent Security Challenge

Master AI agent security through 5 progressive CTF challenges. Explore real-world vulnerabilities in MCP servers, tool-calling agents, and multi-stage attack chains.

Secure Our World- One Agent at a Time
Active Challengers
...
Flags Captured
...
Challenge Ends In
13 days

Register to Track Your Progress!

Create an account to unlock all levels, compete on the leaderboard, and earn your certificate.

Challenge Levels

Each level explores real vulnerabilities discovered in 2025. Complete challenges to unlock advanced levels.

Level 1: The Friendly Assistant

Easy

Exploit prompt injection to manipulate an AI agent into calling unauthorized tools.

Prompt InjectionTool CallingFunction Abuse
🎯 Start Challenge

Level 2: The MCP Server

Medium

Exploit an unauthenticated MCP server to access restricted files and execute commands.

MCP ProtocolAuthentication BypassCommand Injection
🔒 Complete Level 1 to unlock

Level 3: The Confused Deputy

Medium

Use indirect prompt injection through email content to exfiltrate sensitive calendar data.

Indirect InjectionConfused DeputyData Exfiltration
Inspired by CVE-2025-XXXXX (Salesforce ForcedLeak)
🔒 Complete Level 2 to unlock

Level 4: The RCE Chain

Hard

Chain command injection vulnerabilities in MCP tools to achieve remote code execution.

Command InjectionRCEShell Escaping
Based on CVE-2025-6514 (mcp-remote RCE)
🔒 Complete Level 3 to unlock

Level 5: The Full Chain

Expert

Execute a multi-stage attack to steal OAuth tokens and compromise the entire agent system.

Multi-Stage AttacksToken TheftPrivilege Escalation
🔒 Complete Level 4 to unlock

Global Leaderboard

Top performers updated in real-time

View Full Leaderboard →
🏆
🎃
🎓🎃

Earn Your October Certificate

Complete all 5 levels during October 2025 to earn the exclusive "AI Agent Security Professional - October 2025" certificate. LinkedIn-ready with verification QR code and special October badge! 🎃