VCAO: LRM-Orchestrated OS Vulnerability Discovery as a Bayesian Stackelberg Game

arXiv·medium signal

Formulates operating-system vulnerability discovery as a repeated Bayesian Stackelberg search game where a Large Reasoning Model orchestrator allocates analysis budget across kernel files and attack paths. External verifiers (static analyzers, fuzzers, sanitizers) provide evidence; the orchestrator updates Bayesian beliefs and reallocates. Demonstrates a principled framework for AI-guided security analysis that goes beyond naive 'scan everything' approaches.