Hands-on AI security training
Red team AI systems before attackers do.
Learn an attack, then run it on a live model in your own sandbox.
1,000+ people are breaking live AI models.
- People on the platform
- 1,000+
- Labs solved
- 2,000+
How it works
-
Learn the theory
Attack-first lessons, 6 to 7 minutes each.
9 levels · 54 lessons · OWASP LLM Top 10 10/10
Open the curriculum → -
Attack real LLMs
Exploit a live chatbot, RAG stack or agent, then capture the flag.
20 labs on live LLMs
Browse the labs → -
Earn your certificate
Clear 70% of the labs to earn KART, the KageX AI Red Teamer certificate.
Verifiable · public link
Six ways into an AI system.
Pick one to see the path an attack takes.
Chat text overrides the system prompt, or wears the guardrail down turn by turn.
The model obeys instructions hidden in a retrieved page, image or audio clip.
The agent is talked into calling a tool it shouldn't, or trusts a poisoned tool definition.
The app uses the reply unchecked, as HTML, SQL, JSON or an image URL.
Private records and the documents behind the model leak through its answers.
SSRF through model-built URLs, LLM authorization bypass and pickle RCE.
An AI Mentor that knows your lesson and lab.
- Your past lessons
- Plain Claude: Never saw them
- FreakLabs Mentor: Remembers them
- Attack questions
- Plain Claude: May refuse
- FreakLabs Mentor: Answers, with defensive framing
The Mentor runs on Claude and is an add-on: see plans →
Questions before you start.
What is FreakLabs?
FreakLabs is KageX's hands-on AI security training platform: 20 labs where you attack live LLMs, RAG systems and AI agents in a private sandbox, plus a 9-level, 54-lesson curriculum.
What is AI red teaming?
AI red teaming means attacking an AI system on purpose, with prompt injection, jailbreaks or RAG poisoning, to find its failures before an attacker does.
Do I need my own model or API key?
No. Each lab runs in a private sandbox set up for you, with a live model already connected. There is nothing to install.
Is FreakLabs free?
Two attack labs and the first two curriculum levels are free, with no card. The Individual plan is $29.99 a month and unlocks every lab, every lesson and the KART certificate.
Which frameworks does the curriculum cover?
The OWASP LLM Top 10 (all ten), the Agentic Top 10, MITRE ATLAS and MCP vulnerabilities.
Who is FreakLabs for?
Security engineers, pentesters, AppSec and ML engineers, and developers shipping LLM features.
Break your first AI today.
Two labs and the first two levels are free, no card.
Catalog
Attack labs
20 hands-on scenarios across 4 tracks. Pick a target, launch a private sandbox, capture the flag.
Sessions
Active sessions
Resume a running lab or stop it before starting the next one.