Toward Practical RL for Pen‐Testing
Abdul Rahman, Christopher Redino, Dhruv Nandakumar, Tyler Cody, Sachin Shetty, Dan Radke · 2024
Previous chapters motivate reinforcement learning (RL) as a method of simulation in general and in particular how it can be applied to the domain of cybersecurity for penetration testing, but there is still a large leap from this conceptual framework described here and a model that is pragmatically useable. This text and the related works by the authors are far from the first attempt at employing RL for penetration testing, and as previously alluded to, the two main difficulties these works have faced in the past are in achieving scale and realism . In this chapter, we will discuss the current state of these two main issues as they arise in related works, before describing our methods for addressing them.