A custom OpenAI Gym environment that emulates vulnerable networks using Docker containers and virtual machines. It supports:
Deep RL inference takes 50-200ms per decision. In a real pentest, rapid scanning (nmap at 5k packets/sec) produces state updates faster than the agent can process. autopentest-drl
: github.com/autopentest/drl-core (conceptual) A custom OpenAI Gym environment that emulates vulnerable
| Dimension | PentestGPT (LLM) | Autopentest-DRL | | :--- | :--- | :--- | | | Limited by context window | Full state memory | | Exploration strategy | Zero-shot reasoning | ε-greedy, UCB exploration | | Handling unknown exploits | Hallucinates commands | Silent failure (needs reward shaping) | | Cost per episode | High (token-based) | Very low (local compute) | | Best for | Report generation, beginner guidance | Autonomous, high-speed compromise | : github