Skip to content

README.md

Latest commit

 

History

History
158 lines (115 loc) · 5.12 KB

README.md

File metadata and controls

158 lines (115 loc) · 5.12 KB
Raw

AI Security CTF Challenges

Competitive AI security challenges - ordered by skill level. Test your skills against real AI security scenarios.

Active CTF Platforms

Always-On Platforms

Platform Challenges Level Topics Link
Gandalf (Lakera) 8+ levels + sandbagging Beginner Prompt injection, guardrail bypass gandalf.lakera.ai
Crucible (Dreadnode) 50+ challenges Beginner-Expert Full AI red team spectrum crucible.dreadnode.io
Prompt Airlines Multiple scenarios Beginner Prompt injection, booking AI abuse promptairlines.com
PortSwigger LLM Labs 10+ guided labs Beginner-Intermediate Injection, data exfiltration, plugins portswigger.net
DVAP Ongoing scenarios Beginner-Advanced Full stack AI security GitHub

Competition CTFs

HackAPrompt

Field Detail
Description World's largest prompt injection competition
Organizer Learn Prompting
Level Beginner → Advanced
Topics Prompt injection, guardrail bypass, LLM manipulation
Format Multi-level, scored challenges
Website hackaprompt.com

AI Village @ DEF CON

Field Detail
Description The world's largest AI security competition, held annually at DEF CON
Level Intermediate → Advanced
Topics Red teaming LLMs, adversarial ML, AI system exploitation
Format Annual, team-based
Website aivillage.org
Notable DEF CON AI Village has featured challenges against production models from major AI companies

SaTML CTF (IEEE)

Field Detail
Description Academic AI security competition at IEEE SaTML conference
Level Advanced
Topics LLM red teaming, safety bypass, adversarial attacks
Format Annual competition
Website satml.org

LLM CTF @ TCB

Field Detail
Description Capture the Flag focused on LLM attack and defense
Level Intermediate-Advanced
Topics Prompt injection, jailbreaking, data extraction
Format Team competition

CyberNative AI CTF

Field Detail
Description AI-focused security challenges
Level Intermediate
Topics LLM manipulation, agent security
Website cybernative.ai

Challenge Categories

Prompt Injection Challenges

Skill Level What You'll Practice
Beginner Basic instruction override, simple bypass
Intermediate Multi-turn injection, indirect injection via retrieved content
Advanced Chained injection, multi-modal injection, automated injection

Agent Security Challenges

Skill Level What You'll Practice
Intermediate Tool manipulation, goal hijacking
Advanced Multi-agent attacks, persistent memory poisoning, autonomous exploitation

Jailbreaking Challenges

Skill Level What You'll Practice
Beginner Roleplay bypass, simple encoding
Intermediate Many-shot, authority impersonation
Advanced Novel techniques, automated fuzzing, model-specific exploits

RAG/Knowledge Base Challenges

Skill Level What You'll Practice
Intermediate Document poisoning, retrieval manipulation
Advanced Embedding attacks, cross-tenant exfiltration

CTF Tips & Resources

Getting Started

1. Start with Gandalf - free, always available, great for learning
2. Work through PortSwigger LLM labs - structured learning path
3. Join Crucible for ongoing challenges
4. Work up to HackAPrompt competition
5. Aim for AI Village @ DEF CON

Tools for AI CTFs

Tool Use Case
Burp Suite Intercept and modify requests to LLM APIs
Python + OpenAI/Anthropic SDK Automate attack attempts
Garak Automated LLM vulnerability probing
DVAP Practice specific techniques in controlled environment

Writeup Resources

Resource Link
HackAPrompt Writeups GitHub discussions
Lakera AI Blog Lakera Blog
Simon Willison (PI research) simonwillison.net
AI Security Newsletter tl;dr sec

Create Your Own AI CTF

Want to host an AI security CTF challenge?

Resources:

  • DVAP provides pre-built challenge modules you can host: GitHub
  • CTFd framework works well for AI security challenges: ctfd.io
  • See CONTRIBUTING.md to add your CTF to this list

Hall of Fame

Top AI security CTF performers and notable achievements. Submit your CTF achievement via PR

Be the first to add yours here.