Reproduce “PBFuzz: Agentic Directed Fuzzing for PoV Generation” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Haochen Zeng, Andrew Bao, Jiajun Cheng, Chengyu Song

PBFuzz helps generate proof-of-vulnerability inputs for known bugs by having an AI agent infer reachability and triggering conditions, encode them as parameters, and use property-based testing to search the resulting input space. On Magma, it triggered 57 vulnerabilities, or 51 excluding six custom-harness cases, compared with 49 for AFL++ with CmpLog; it assumes known predicates and execution signals.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.