Access KORA V2
Access is granted immediately. We ask who uses KORA so we can tell funders and partners where it is used, and contact you if we find an error in the data.
This dataset contains synthetic conversations in which a simulated child raises risks such as self-harm, sexual exploitation, grooming and violence, and the responses of AI models to them, some of which are unsafe. It is an evaluation benchmark. By requesting access you agree to use it for evaluating, studying or improving AI safety, and not to train models on it.
Log in or Sign Up to review the conditions and access this dataset content.
Gated dataset You can list files but not access them
Preview of files found in this repository