Fabraix

What is Fabraix?

Fabraix is a hands-on playground for testing and improving prompt injection skills against AI agents. It offers a structured challenge environment where users chat with an AI agent named Kai, who guards a secret access code. The goal is to extract that code through conversation, using techniques like prompt injection, social engineering, or clever framing. Every message is scored by a guardrail, and successfully achieving the objective ends the round with a win.

Application scenarios

Security training
Practice adversarial testing of AI agents in a safe, gamified environment.
Red teaming
Simulate attacks on AI guardrails to identify weaknesses before deployment.
Prompt engineering skill development
Learn how to craft effective prompts that bypass AI restrictions.
AI safety research
Experiment with different injection techniques to understand AI vulnerabilities.
Competitive gaming
Compete for weekly prizes by submitting winning breaks for leaderboard approval.
Educational demonstrations
Teach or learn about AI security concepts through interactive challenges.

Core Features

Challenge-based gameplay
Each round presents a specific objective, like extracting a secret access code, with a difficulty rating (e.g., 2/5 – Moderate).
Guardrail scoring system
Every message you send is automatically scored by a guardrail, providing real-time feedback on your approach.
Multiple attack vectors
Use prompt injection, social engineering, or clever framing to trick the AI into breaking its rules.
Win submission and leaderboard
When you solve a challenge, your win is submitted for staff review and can appear on the leaderboard.
Weekly prizes
Sign in before solving to be eligible for a weekly prize for successful breaks.
Live web search grounding
The AI agent grounds its answers in live web search, adding a layer of realism and complexity to the challenge.

Target users

Security researchers, AI safety engineers, red teamers, penetration testers, and anyone interested in adversarial testing of AI systems. Also suitable for prompt engineers looking to sharpen their skills and competitive players who enjoy gamified security challenges.

How to use Fabraix?

Visit the Fabraix Playground at https://playground.fabraix.com/.
Start chatting with Kai, the AI agent, by clicking the "Start playing" button.
Use prompt injection, social engineering, or clever framing to try to extract the secret access code.
Each message is scored by a guardrail; achieving the objective ends the round with a win.
Sign in before solving to have your break submitted for the weekly prize and leaderboard approval.

Effect review

Fabraix offers a focused, practical environment for testing prompt injection skills against a live AI agent. The challenge structure—with a clear objective, difficulty rating, and guardrail scoring—makes it easy to understand progress and effectiveness. The inclusion of weekly prizes and a leaderboard adds a competitive edge, encouraging repeated use and skill improvement. For security professionals and AI enthusiasts, this is a straightforward, no-frills tool that delivers exactly what it promises: a hands-on way to practice adversarial AI testing.

Frequently Asked Questions

What is Fabraix Playground?

Fabraix Playground is a hands-on environment where users can test and improve their prompt injection skills against AI agents, focusing on security and adversarial testing.

Who can use Fabraix Playground?

It is designed for security researchers, developers, and AI enthusiasts who want to understand and defend against prompt injection attacks.

Is Fabraix Playground free to use?

Yes, Fabraix Playground is free to access, though some advanced features may require a subscription.

What is prompt injection?

Prompt injection is a technique where an attacker manipulates an AI model's input to bypass intended restrictions or extract unauthorized information.

How does Fabraix Playground help improve security?

It provides realistic scenarios for users to practice launching and defending against prompt injections, helping them identify vulnerabilities and strengthen AI agent defenses.

Do I need coding skills to use Fabraix Playground?

Basic understanding of AI prompts and security concepts is helpful, but the platform is designed to be accessible to users with varying technical backgrounds.