Automated Penetration Testing Based on LSTM and Advanced Curiosity Exploration
Qiankun Ren, Jingju Liu, Xinli Xiong, Canju Lu · 2024
With the rapid development of artificial intelligence technology, cybersecurity issues have become increasingly prominent, particularly the escalating threats posed by malicious cyber behavior to users and systems. Automated penetration testing is increasingly important as a crucial means of maintaining cybersecurity. However, existing automated penetration testing methods based on reinforcement learning algorithms exhibit deficiencies in leveraging historical experience information and exploring unknown environments. This paper proposes an innovative algorithm—PLACE (PPO algorithm based on LSTM architecture and advanced curiosity exploration mechanism) to address these issues. By incorporating an LSTM architecture and an advanced curiosity module, this algorithm enhances the efficiency of utilizing historical experience information and improves the exploration capability in unknown environments. The experimental results show that the cumulative reward of the PLACE algorithm in the penetration test scenario composed of 40 nodes is 13.84\% higher than that of the CPPO algorithm, 12.95\% higher than that of the PPO-Curious-advanced algorithm, and 13.29\% higher than that of PPO algorithm. Cumulative rewards and success rates in other network scale scenarios have also improved, providing new insights and approaches to solving complex problems in the field of cybersecurity.