LLM-Driven Vulnerability Discovery Through Bugonomics Lens

An arXiv paper (2605.24632) recently delves into the discovery of vulnerabilities driven by large language models (LLMs) through the perspective of bugonomics, emphasizing the economic aspects of creating, validating, prioritizing, and addressing security-related flaws. Traditionally, high-end bugonomics was focused on offensive pricing, with costly production-grade zero-days and exploit chains being specialized outputs for governments, brokers, and offensive vendors. Meanwhile, defender-side bugonomics has been present in vulnerability research, reward initiatives, and vendor remediation efforts. The introduction of LLM-assisted systems enhances the accessibility of generating candidates, understanding code, constructing harnesses, drafting proof-of-impact, and writing reports. The paper contends that while media often highlights AI's potential to transform security, it seldom examines the associated costs and incentives.