STAF: Leveraging LLMs for Automated Attack Tree-Based Security Test Generation