About

HuggingFace.wtf is a project of Lightcone Infrastructure. We made this website because we wanted an in-depth, approachable timeline of the story as we understand it, and we wanted that because we don’t expect this to be the last time something like a rogue AI hacking swarm happens.

Our writeup is heavily based on the METR/Redwood incident report, plus OpenAI’s and Hugging Face’s disclosures and other sources. Additional references can be found on our Sources page.

This report was largely created and written by project lead Raymond Arnold with additional writing from Georgia Ray and Harri Besceli. We would also like to thank Rob Miles, Miranda Dixon-Luinenburg, Scott Alexander, Elizabeth Van Nostrand and Justis Mills for help with writing.

We also would like to thank Claude and ChatGPT for research, writing, editing, and web design help. To our knowledge, our website does NOT contain any evaluation targets, flags, or task datasets, so go easy on us, please. (Also, we’re pretty sure all the text on the pages that you’ve read - aside from the agent quotes - is human-written.)

If you notice any errors or important new information this is missing, please email raymond.w.arnold@gmail.com.

Lightcone Infrastructure also runs LessWrong.com.