Multi-Agent Reinforcement Learning for Red Teaming & Penetration Testing

A final-year project implementing Multi-Agent Deep Deterministic Policy Gradient (MADDPG)
for simulating red-team penetration testing strategies in a cybersecurity environment.

---

📌 Overview

This project simulates a red-team penetration testing scenario using Multi-Agent Reinforcement Learning (MARL).
Agents represent attackers attempting to discover and exploit vulnerabilities in a simulated computer network,
with the ultimate goal of compromising a designated target node.

---

✨ Features

• Multi-Agent environment for red-team style attack simulations.
• Cybersecurity network modeled as a graph with nodes and services.
• Implementation of MADDPG (Multi-Agent Deep Deterministic Policy Gradient).
• Training and evaluation scripts with visualization support.
• Custom Replay Buffer for experience replay.

---

📂 Project Structure

├── training.py # Main training loop

├── evaluate.py # Run trained agents

├── cyber_env.py # Custom cybersecurity environment

├── maddpg.py # MADDPG agent and neural networks

├── requirements.txt # Python dependencies

└── README.yaml # Project documentation

---

⚙️ Installation

1. Clone the repository:

   git clone https://github.com/your-username/MARL-RedTeam.git
   cd MARL-RedTeam

2. Create a virtual environment:

python3 -m venv venv
source venv/bin/activate   # Linux/Mac
venv\Scripts\activate      # Windows

3. Install dependencies:

pip install -r requirements.txt

🚀 Training

Run the training script:

python training.py

Progress is displayed every 100 episodes with:

• Average reward
• Success rate

🧪 Evaluation

• After training, save agent weights
• Evaluate trained agents:

  python evaluate.py

📊 Visualization

During training, results are plotted:

• Episode Rewards
• Success Rate (rolling average)

The environment can also render:

• Node states
• Compromised systems

📦 Requirements

• Python 3.10+
• PyTorch
• NumPy
• NetworkX
• Matplotlib

🔮 Future Work

• More realistic vulnerability models
• Blue-team (defender) agents -Larger, dynamic network topologies Transfer learning for different environments

📜 License

This project is licensed under the MIT License.