Grey-Box Fuzzing Based on Reinforcement Learning for XSS Vulnerabilities

Cross-site scripting (XSS) vulnerabilities are significant threats to web applications. The number of XSS reported has increased annually for the past three years, posing a considerable challenge application maintainers. Black-box scanners mainstream tools security engineers perform penetration testing and detect vulnerabilities. Unfortunately, black-box rely on crawlers find input points applications cannot guarantee all tested. To this end, we propose grey-box fuzzing method based reinforcement learning, which can reflected stored Java We first use static analysis identify potential from components (i.e., code, configuration files, HTML files) application. Then, an vulnerability payload generation is proposed, used together with learning model. define state, action, reward functions models detection scenarios so that fuzz loop be performed automatically. demonstrate effectiveness proposed method, compare it against four state-of-the-art scanners. Experimental results show our finds no false positives.