This research introduces the GAP framework (Generalizable Autonomous Pentesting) that combines domain randomization with meta-reinforcement learning to create more effective AI security systems.

• Addresses the training environment dilemma where simulated training lacks real-world relevance
• Implements domain randomization to expose AI agents to diverse security scenarios
• Uses meta-reinforcement learning to help AI quickly adapt to new environments
• Demonstrates improved generalization capability across varying network security contexts

This innovation matters because it brings autonomous penetration testing closer to practical deployment in real-world cybersecurity operations, potentially reducing human workload while improving vulnerability detection.

Mind the Gap: Towards Generalizable Autonomous Penetration Testing via Domain Randomization and Meta-Reinforcement Learning