What AI says about Unsupervised Bias Detection

NGO-Algorithm-Audit's unsupervised-bias-detection is a tool designed to detect biases in binary AI classifiers without requiring protected attribute data. It combines data-driven statistical analysis with qualitative, human expert judgment to identify complex, intersectional, and disparate treatment across groups, and it is model-agnostic. The project provides open-source code, an audit workflow, and a web app and accompanying scientific paper to apply real-world audits and generate audit reports.