Welcome to Haize Labs

Check out our new paper!

Endless Jailbreaks with Bijection Learning: a Powerful, Scale-Agnostic Attack Method finds jailbreaks by teaching language models new languages. Best of all, these languages are programatically generated, allowing any attack to be encoded in any language. Read more on our blog.

Platform Overview

Haize Labs develops automated red teaming, filtering, and monitoring tools.

Detectors Detectors catch and understand the content you care about.

Automated Red Teaming Continously monitor model performance and safety at lightning fast speeds.

Frequently asked questions Have a question? This page might answer it.