VCAO: Verifier-Centered Agentic Orchestration for Strategic OS Vulnerability Discovery

TL;DR

VCAO introduces a game-theoretic framework using Bayesian reasoning and multiple analysis tools to improve Linux kernel vulnerability discovery efficiency and accuracy.

Contribution

It formulates OS vulnerability discovery as a Bayesian Stackelberg game and develops VCAO, a novel architecture with formal budget allocation and verification strategies.

Findings

VCAO discovers 2.7x more vulnerabilities per budget than fuzzing.

It outperforms static analysis and non-game-theoretic pipelines in vulnerability detection.

Reduces false positives by 68% compared to baseline methods.

Abstract

We formulate operating-system vulnerability discovery as a \emph{repeated Bayesian Stackelberg search game} in which a Large Reasoning Model (LRM) orchestrator allocates analysis budget across kernel files, functions, and attack paths while external verifiers -- static analyzers, fuzzers, and sanitizers -- provide evidence. At each round, the orchestrator selects a target component, an analysis method, and a time budget; observes tool outputs; updates Bayesian beliefs over latent vulnerability states; and re-solves the game to minimize the strategic attacker's expected payoff. We introduce \textsc{VCAO} (\textbf{V}erifier-\textbf{C}entered \textbf{A}gentic \textbf{O}rchestration), a six-layer architecture comprising surface mapping, intra-kernel attack-graph construction, game-theoretic file/function ranking, parallel executor agents, cascaded verification, and a safety governor. Our DOBSS-derived MILP allocates budget optimally across heterogeneous analysis tools under resource constraints, with formal $\tilde{O}(\sqrt{T})$ regret bounds from online Stackelberg learning. Experiments on five Linux kernel subsystems -- replaying 847 historical CVEs and running live discovery on upstream snapshots -- show that \textsc{VCAO} discovers $2.7\times$ more validated vulnerabilities per unit budget than coverage-only fuzzing, $1.9\times$ more than static-analysis-only baselines, and $1.4\times$ more than non-game-theoretic multi-agent pipelines, while reducing false-positive rates reaching human reviewers by 68\%. We release our simulation framework, synthetic attack-graph generator, and evaluation harness as open-source artifacts.