HarmBench
Center for AI SafetyA standardized evaluation framework for automated red-teaming from the Center for AI Safety. It runs a common pipeline across many attack methods and target models so red-team attacks and defenses can be compared rigorously and co-developed.
HarmBench, MIT, Center for AI Safety; ~1K GitHub stars (June 2026). Standardized red-teaming eval across 18 attacks and 33 target models. Verified live June 2026.
Openness
5 high confidence- license
- MIT(OSI)
- source
- public
- self-host
- yes
- service
- none
- core-gated
- ungated
Fully open source: MIT-licensed standardized red-teaming evaluation framework, self-hostable.
- https://github.com/centerforaisafety/HarmBench recorded 2026-06-29
MIT; standardized automated red-teaming eval; 18 attacks x 33 target models; ~993 stars
Adoption
2 low confidence~1K GitHub stars (June 2026); stars-only proxy. Widely cited as the standardized red-teaming benchmark in research.
- https://github.com/centerforaisafety/HarmBench recorded 2026-06-29
~993 stars (June 2026)
Capability
4 medium confidenceThe reference standardized harness for comparing red-team attacks and defenses; research-grade breadth.
- https://github.com/centerforaisafety/HarmBench recorded 2026-06-29
18 attack methods, 33 target models, standardized pipeline
Unchanged since 2026-08-01 (last edited, not re-checked)