HarmBench

What it is

A standardized evaluation framework for automated red teaming of large language models, measuring how often attacks get models to comply with harmful requests and how reliably models refuse them.

Topics