Turn your research into a
publication-ready draft

The rapid advancement and widespread adoption of Large Language Models (LLMs) have significantly transformed various domains, including software development [1, 2]. These AI-powered models are increasingly capable of generating functional code snippets, entire functions, and even complex application architectures, promising to accelerate development cycles and enhance programmer productivity [1, 2]. However, this burgeoning reliance on AI for code generation introduces a critical and evolving challenge: ensuring the security of the output. Web applications, in particular, remain a primary target for malicious actors due to the sensitive data they often manage and their broad accessibility across public networks [3, 4]. The inherent complexity of modern information systems further amplifies these security concerns, making robust vulnerability assessment and mitigation paramount [3].

The security landscape of web applications has long been guided by established frameworks and methodologies. The Open Web Application Security Project (OWASP) has been instrumental in this regard, providing widely recognized lists of common application security risks, most notably the OWASP Top Ten [4, 5]. These guidelines serve as a critical benchmark for developers and security professionals to identify, understand, and address prevalent vulnerabilities. Concurrently, the field of web application security benchmarking has seen numerous proposals and efforts to systematically evaluate security tools and practices [6, 7].