The AI Model Risk Catalog: What Developers and Researchers Miss About Real-World AI Harms

TL;DR

This paper introduces the AI Model Risk Catalog, derived from analyzing nearly 460,000 model cards, highlighting gaps in risk reporting by developers and researchers, especially regarding social and systemic harms.

Contribution

The paper presents a comprehensive catalog of AI risks and compares developer and researcher focus, emphasizing the need for structured risk reporting to address overlooked harms.

Findings

Developers focus on bias and safety, while researchers emphasize social impacts.

Both groups overlook fraud and manipulation risks.

Structured risk reporting can improve early identification of systemic harms.

Abstract

We analyzed nearly 460,000 AI model cards from Hugging Face to examine how developers report risks. From these, we extracted around 3,000 unique risk mentions and built the \emph{AI Model Risk Catalog}. We compared these with risks identified by researchers in the MIT Risk Repository and with real-world incidents from the AI Incident Database. Developers focused on technical issues like bias and safety, while researchers emphasized broader social impacts. Both groups paid little attention to fraud and manipulation, which are common harms arising from how people interact with AI. Our findings show the need for clearer, structured risk reporting that helps developers think about human-interaction and systemic risks early in the design process. The catalog and paper appendix are available at: https://social-dynamics.net/ai-risks/catalog.