DEF CON Red Teaming Dataset

This is the dataset from the 2023 DEF CON AI Village generative AI red teaming event, organized by Humane Intelligence. Thousands of participants tried to get eight models to misbehave across challenges like producing misinformation, revealing hidden prompts, or making discriminatory statements. The released data includes the conversations and which attempts succeeded.

It's messy and human, which is its value: these are the things ordinary people actually try.