Project

BinaryAudit

BinaryAudit is a benchmark of AI agents looking for backdoors in compiled binaries of open-source software. Agents receive stripped binaries and Ghidra or Radare2 inside Docker. The task set covers lighttpd, dnsmasq, Dropbear, Sozu, and Caddy, including clean binaries and timebomb tasks. The README says the backdoors were artificially injected.

Targets: Binary
Category: Native analysis tools
Relationship: Foundation tool
Sources reviewed: 2026-11-07

Stars: 107
Last commit: 2026-10-22
Language: Shell

GitHub data as of 2026-11-09

Copy as Markdown View JSON Cite

Requirements & limits

Harbor, installed with uv tool install harbor, plus Docker and an API key for the chosen model.
Before Docker tasks, build the image binaryaudit-base:latest from docker/base.Dockerfile.
License: Apache 2.0.

Cite

BibTeX

@misc{binaryaudit,
  title = {{BinaryAudit}},
  howpublished = {\url{https://github.com/QuesmaOrg/BinaryAudit}},
  url = {https://github.com/QuesmaOrg/BinaryAudit},
  urldate = {2026-11-07},
  note = {Awesome REA entry: \url{https://awesomerea.org/resources/binaryaudit/}}
}

Plain text

BinaryAudit. https://github.com/QuesmaOrg/BinaryAudit (accessed 2026-11-07). Awesome REA entry: https://awesomerea.org/resources/binaryaudit/

Useful? Star Awesome REA on GitHub2.4k

About this entry

The description was reviewed against primary sources.

Report an outdated entry Suggest a translation fix

Primary sources

README.md GitHub · QuesmaOrg/BinaryAudit · Revision 175b7ec

More in Native binaries

All 502 entries in Native binaries