The AI Incident Database
The AI Incident Database is an open source repository of real-world incidents where AI systems caused—or nearly caused—harm. A project of the Responsible AI Collaborative, the database is a useful resource cataloging more than 1200 incidents, including deepfake investment and phishing scams, data leaks, and autopilot car collisions.
A number of features built into the site can help researchers and AI adopters utilize and even submit to the data:
- A “Discover” interface where users can browse incidents in list, table or spatial visualization.
- A “Submit Incident Reports” function inviting crowdsourced contributions from developers, researchers and the public.
- Taxonomies and classification tools to sort incidents by type, severity, domain and other dimensions.
- Downloadable data and research-friendly formats.
As AI enters more and more dimensions of work and life, the database helps raise awareness of the sort of risks stakeholders should consider as they explore AI integrations. “Like similar databases in aviation and computer security, the AI Incident Database aims to learn from experience so we can prevent or mitigate bad outcomes,” the collaborative explains.