Scholarlink

With the fast development of Internet connection and 6G technology, Penetration testing is gained high interest in identifying effective attacks in large scale networks. Traditional methods often rely on many searches’ techniques, which can be computationally too much and inefficient in dynamic network environments. This paper presents a Reinforcement Learning based framework for performing automatic penetration testing. The proposed framework automated attack path discovery, Showing Rainbow Deep Q Networks (DQN) to optimize the decision-making process in penetration testing scenarios. The proposed framework represents network vulnerabilities as an attack graph and shows the attack path discovery problem as a Markov Decision Process (MDP), where an agent navigates the graph to uncover critical attack paths. Experimental evaluations demonstrate the effectiveness of this approach in optimizing attack path discovery, enhancing explore efficiency, and decision-making stability.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Reinforcement Learning-Based Framework for Automated Penetration Testing

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Muhammed Karam Fathi,
  • Khaled Metwally,
  • Mohamed Sobh,
  • Ayman Bahaa Eldin

摘要

With the fast development of Internet connection and 6G technology, Penetration testing is gained high interest in identifying effective attacks in large scale networks. Traditional methods often rely on many searches’ techniques, which can be computationally too much and inefficient in dynamic network environments. This paper presents a Reinforcement Learning based framework for performing automatic penetration testing. The proposed framework automated attack path discovery, Showing Rainbow Deep Q Networks (DQN) to optimize the decision-making process in penetration testing scenarios. The proposed framework represents network vulnerabilities as an attack graph and shows the attack path discovery problem as a Markov Decision Process (MDP), where an agent navigates the graph to uncover critical attack paths. Experimental evaluations demonstrate the effectiveness of this approach in optimizing attack path discovery, enhancing explore efficiency, and decision-making stability.