AI engineering glossary · Evals guardrails AI security
What is harmBench?
HarmBench: A standardised benchmark and framework for automated red teaming and refusal testing of language models, providing a set of harmful behaviours, attack methods and a classifier to judge outputs so that attacks and defences can be compared fairly.
HarmBench: an example
A lab runs several attack methods from a standard suite against its model and compares attack success rates with a rival's.
Related terms
More Evals guardrails AI security terms
- HELM
- HumanEval
- Improper output handling
- Input guardrail
- Intrinsic hallucination
- ISO/IEC 42001
- Jailbreak
- Judge calibration
Where you learn it
HarmBench is part of the AI Engineering Program, which teaches it with notes, a hands-on lab and an exam. The first lesson is free.