This got me wondering how easy it'd be to automate discovery of BYOVD vulns with LLMs (both offensively and defensively)