Reproduce “Where AI Vulnerability Discovery Harnesses Go Quiet: Structural False Negatives and Hidden Vulnerabilities” 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

Gadi Evron

The authors map 13 mechanisms by which 12 open-source security-scanning harnesses can omit vulnerable code or findings. They examine selection, parsing, analysis, verification, and reporting stages, then test file-selection rules in six harnesses without running models. In a planted fixture, all six dropped the code before analysis; five did so without notice, showing possibility, not prevalence.

Summary by alphaXiv.

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