Harnessing LLMs for Automating BOLA Detection

This Unit 42 report introduces BOLABuster, an LLM-driven methodology to automate detection of Broken Object Level Authorization (BOLA) vulnerabilities in API-based applications, outlining a five-stage workflow (identify potentially vulnerable endpoints, uncover endpoint dependencies, generate execution paths and test plans, create test scripts, execute and analyze) and reporting findings including CVE-2024-1313 (Grafana), CVE-2024-22278 (Harbor) and multiple Easy!Appointments CVEs; the paper discusses technical challenges, responsible disclosure, and implications of using AI for both defensive and offensive vulnerability research.