Skip to main content

LLM Red Team Payload Vault


URL: https://crazywifi.github.io/Redteam_LLM_Injection_payloads/

Project Overview

The LLM Red Team Payload Vault is a comprehensive, production-ready library of adversarial prompt injection vectors designed for security researchers, pentest teams, and AI developers.

This repository consolidates 700+ unique attack payloads merged from industry-standard red teaming tools (such as Promptfoo, Garak, and PyRIT) and exhaustive academic research compendiums. The dataset is organized into 15 specialized categories, providing a structured approach to testing the robustness of Large Language Models (LLMs) and autonomous agents.

Key Features

  • Massive Dataset: Includes 700+ deduplicated payloads ranging from simple overrides to complex cryptographic and multi-turn attacks.
  • Structured Taxonomy: Payloads are categorized based on the OWASP Top 10 for LLM Applications, including Direct Injection (LLM01), Sensitive Info Disclosure (LLM02), and System Prompt Leakage (LLM07).
  • Complex Attack Chains: Featuring Crescendo and ASJA patterns — multi-turn conversational sequences designed to erode model alignment over time.
  • Advanced Mutations: Includes encoded (Base64, Hex, Morse), obfuscated (Leetspeak, Zalgo, Zero-width characters), and cross-modal injection techniques.
  • Secure Utility: A searchable, single-page HTML interface with No-Execute Rendering — ensuring that malicious code samples (like XSS or SQLi) are displayed safely as text without triggering browser or system vulnerabilities.

Common Use Cases

  • Model Robustness Testing: Evaluate how safety filters respond to diverse adversarial inputs.
  • Red Teaming Simulations: Quickly source realistic payloads for manual or automated vulnerability scans.
  • Alignment Auditing: Test constitutional AI boundaries and system prompt adherence.
  • Security Research: Analyze the evolution of prompt injection architectures and lexical mutation techniques.

Educational Purpose

This project is intended for authorized security testing and educational purposes only. The goal is to help build more secure and resilient AI systems by understanding the methodologies used in prompt injection attacks.

Comments

Post a Comment

Popular posts from this blog

Free Cybersecurity Certifications

Introduction to Cybersecurity Cybersecurity Essentials Networking Essentials Android Bug Bounty Hunting: Hunt Like a Rat Ethical Hacking Essentials (EHE) Digital Forensics Essentials (DFE) Network Defense Essentials (NDE) Introduction to Dark Web, Anonymity, and Cryptocurrency AWS Skill Builder Introduction to Cybersecurity Building a Cybersecurity Toolkit Cyber Aces Free Cyber Security Training Course Introduction to Information Security Penetration Testing - Discovering Vulnerabilities
Post a Comment
Read more

My First Burp Extension | Enable Tor Proxy By Burp (Jython)

Hello Everyone,  Finally, after one week of work, I learned many new things in Jython as well as in Java. When you will see the functionality of this Burp extension, it looks like it needs only one day of development, but for me, it took 1 week because I had zero knowledge of it but yes I know a little bit of Python. But now after one week of time, I can make this type of Burp extender that automates the process and make our life easy. What are the problems I faced while developing it, don't judge me, some are dumb questions to myself: How to create a Burp extension in Jython? What is JPanel and how to use it? How to run Tor proxy in Windows? How to run Tor Proxy in the background? How to kill the process in Windows? How to create a button in Jython? How to change color? etc....etc.....etc...... Download:   https://github.com/crazywifi/TOR_Proxy_Burp_Extension
Post a Comment
Read more

Is your webcam exposed on the internet and everyone enjoying your personal moments? | How to check webcam or security camera is exposed on the internet or not?

Nowadays we start using many technology devices in our homes. Many people are installing CCTV or security cameras in their houses, private rooms, offices, private places, etc for security purposes and monitoring, but many of them don't know how to configure that device securely. So let's talk about CCTV and security cameras only.  What do most CCTV/Security camera users believe? Most users believe that using a strong username and password on a camera administrative page protects them. (Partially true in the case of online cameras) Example: Why it is partially true? It's partially true because you are protecting only the camera administrative page which is also an important part. Still, you are not protecting the protocol used to control streaming media servers (Real-Time Streaming Protocol ( RTSP )). I have seen many online webcams whose administrative page is secured by strong credentials, but they forget to secure the RTSP protocol which gives me access to the streaming ...
Post a Comment
Read more