Automated Penetration Testing Based on MATD Algorithm

With the rapid advancement of internet technology, cybersecurity risks are continuously increasing, and malicious cyberattacks have caused significant losses to users, making it crucial to maintain network security. Automated penetration testing plays a vital role in current cybersecurity efforts. However, existing automated penetration testing methods suffer from inefficiencies in information utilization, inaccurate action decision-making, and an overly large action space. To address these issues, we propose a new algorithm called Multi-Agent Task Decomposition(MATD), based on historical experience information and task decomposition. MATD enhances the efficiency of utilizing historical observation data through the use of Long Short-Term Memory(LSTM) networks and reduces the action space of individual agents by employing a multi-task decomposition method. Experimental results demonstrate that the proposed method outperforms the traditional Proximal Policy Optimization(PPO) algorithm by achieving higher cumulative rewards in fewer steps, thereby improving decision-making in intelligent penetration testing. This approach holds promise for increasing the efficiency of penetration testing and enhancing overall network security.