Introduction As Large Language Models become a core part of modern applications, they introduce an entirely new class of security risks. Vulnerabilities such as prompt injection, system prompt leakage, data poisoning, excessive agency, and insecure tool usage cannot be fully understood by reading documentation alone—they need to be experienced in practice.
LLMVault is an open-source, intentionally vulnerable training platform designed to teach the OWASP Top 10 for LLM Applications (2025) through hands-on exploitation. Inspired by projects like WebGoat and KubeGoat, LLMVault provides a safe, self-hosted environment where learners attack vulnerable AI assistants, capture flags, and immediately learn the corresponding defensive techniques.
The platform follows a CTF-style approach with progressively challenging labs covering real-world LLM attack scenarios. Every challenge is paired with mitigation guidance, helping users understand not only how vulnerabilities are exploited, but also how to prevent them in production systems.
Designed for security professionals, developers, students, and AI practitioners, LLMVault requires no API keys, runs fully offline, supports Docker deployment, and offers reproducible labs that can be completed without cloud dependencies.
Whether you're beginning your AI security journey or preparing to secure enterprise LLM applications, LLMVault provides a practical environment to learn by breaking vulnerable systems first.
Key Features : Hands-on labs aligned with the OWASP Top 10 for LLM Applications (2025) CTF-style challenges with progressive difficulty Learn both offensive techniques and defensive mitigations Fully offline with no API keys or cloud costs One-command deployment using Docker or Python Extensible architecture for creating custom labs Scoreboard, hints, and completion tracking
Quick Start
# Clone the repository
git clone https://github.com/CyberSunil/LLMVault.git
# Enter the project
cd LLMVault
# Install dependencies
pip install -r requirements.txt
# Start the application
python app.py
# Open in your browser
http://127.0.0.1:5000
GitHub Repository: github.com/CyberSunil/LLMVault
