This repository contains documentation and explanation for CSET's harm taxonomy of incidents from AIID.
The AI Incident Database is a community-driven AI incident reporting site. Users are "invited to explore the incidents collected to date, view the complete listing, and submit additional incident reports. Researchers are invited to review our working definition of AI incidents."
Between June 2022 and June 2024, CSET developed and applied an AI Harm incident taxonomy to nearly 300 incidents included in the AI Incident Database. Annotation work has been discontinued, but the team is still accessible for any questions regarding the taxonomy and potential use cases.
Documentation for CSET's annotation process can be found here. This schema includes the motivation for annotated fields, along with a data dictionary for the fields for assistance in data processing.