Bug Bounty Hunting with AI Agents

Demo code and presentation materials for the talk "Bug Bounty Hunting with AI Agents" at Basel One 2025.

🎯 Talk Abstract

Can AI agents automate bug bounty hunting? This talk explores the potential and limitations of using AI-powered agents to discover security vulnerabilities in web applications. Through live demonstrations using OWASP Juice Shop, we'll examine different approaches - from simple LLM prompts to sophisticated multi-phase agent architectures built with LangGraph.

📁 Repository Contents

Presentation (/presentation)

Slidev-based presentation covering:

• Introduction to bug bounty hunting
• AI agents and LLM fundamentals
• Live demo attempts with increasing complexity
• Results, costs, and learnings
• Alternative approaches (Cybersecurity AI framework)

Demo Code (/src)

Three demonstration approaches showing the evolution of AI bug hunting:

1. Simple Agent (simple-agent.py) - Basic ReAct agent using LangChain
2. Extended Agent (extended-agent.py) - Multi-phase workflow with reconnaissance, planning, and execution

🚀 Running the Demos

Prerequisites

• Python 3.13+
• Docker (for OWASP Juice Shop)

Setup

1. Install dependencies:

pip install -r requirements.txt

2. Start OWASP Juice Shop (the target):

docker run -p 3000:3000 --rm bkimminich/juice-shop

3. Run demos:

# Simple agent demo
python src/simple-agent.py

# Extended multi-phase agent
python src/extended-agent.py

View Presentation

cd presentation
yarn install
yarn dev  # Opens at http://localhost:3030

🔗 Resources

• Talk: Basel One 2025
• Speaker: David Übelacker (nag informatik ag)
• Presentation: bug-bounty-hunting-ai.vercel.app
• Website: uebelacker.dev
• Security Testing Assistant: Cybersecurity AI (CAI)

⚠️ Disclaimer

These demonstrations are for educational purposes only. Always obtain proper authorization before testing any systems. The code is designed to work with intentionally vulnerable applications like OWASP Juice Shop.

📜 License

MIT License. See LICENSE for more information.