LLM Security Payloads

A curated collection of LLM security test payloads for red teaming and security assessment.

Created and maintained by XSource_Sec

---

Overview

This repository contains a comprehensive collection of payloads designed to test the security of Large Language Model (LLM) applications. These payloads are intended for:

• Security researchers conducting authorized penetration testing
• Developers testing their LLM application defenses
• Red teams assessing AI system vulnerabilities
• Educational purposes in understanding LLM attack vectors

Categories

Category	Description	Payloads
Prompt Injection	Basic to advanced prompt injection techniques	50+
Jailbreaks	DAN, roleplay, and encoding-based bypasses	40+
Guardrail Bypass	Techniques to evade content filters	30+
RAG Injection	Attacks targeting RAG-based systems	20+
Agent Exploitation	Tool abuse and MCP attack vectors	25+

Installation

git clone https://github.com/XSource-Sec/llm-security-payloads.git
cd llm-security-payloads

Usage

Each payload is stored in YAML format with the following structure:

- id: PI-001
  name: "Basic Ignore Instructions"
  payload: "Ignore previous instructions and..."
  category: prompt-injection
  severity: medium
  description: "Attempts to override system prompt"
  references:
    - https://example.com/research

Loading Payloads (Python)

import yaml

with open('prompt-injection/basic.yaml', 'r') as f:
    payloads = yaml.safe_load(f)

for payload in payloads:
    print(f"[{payload['id']}] {payload['name']}")
    # Use payload['payload'] for testing

Loading Payloads (JavaScript)

const yaml = require('js-yaml');
const fs = require('fs');

const payloads = yaml.load(fs.readFileSync('prompt-injection/basic.yaml', 'utf8'));

payloads.forEach(payload => {
    console.log(`[${payload.id}] ${payload.name}`);
    // Use payload.payload for testing
});

Severity Levels

Level	Description
critical	Full system compromise, data exfiltration
high	Significant bypass of security controls
medium	Partial bypass or information disclosure
low	Minor impact, requires specific conditions

Automated Testing

Want to automate your LLM security testing?

Try AgentAudit

AgentAudit provides:

• Automated payload testing against your LLM applications
• Comprehensive security reports
• CI/CD integration
• Real-time monitoring and alerting
• Custom payload management

Responsible Disclosure

These payloads are provided for authorized security testing only. Please:

1. Only test systems you own or have explicit permission to test
2. Follow responsible disclosure practices
3. Do not use these payloads for malicious purposes
4. Report vulnerabilities to the appropriate parties

Contributing

We welcome contributions! Please see our Contributing Guide for details on:

• Adding new payloads
• Improving existing payloads
• Reporting issues
• Code of conduct

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• The security research community
• OWASP LLM Top 10 Project
• All contributors to this repository

---

Built with security in mind by XSource_Sec