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

2025-08-21

AI Summary

AI developers and researchers disproportionately emphasize technical risks in risk reporting, while systematically overlooking human-interaction–driven harms—such as fraud and manipulation—that manifest in real-world deployments. Method: We conduct the first large-scale comparative analysis of 460,000 Hugging Face model cards, the MIT AI Risk Repository, and the AI Incident Database, employing NLP-driven risk classification and manual coding to construct a publicly available, empirically grounded AI model risk taxonomy comprising 3,000 distinct risk entries. Contribution/Results: Our analysis reveals a significant misalignment between developer/researcher risk priorities and the actual distribution of documented AI incidents—particularly underestimating socio-technical and interactive risks. We therefore propose integrating structured, human-centered, and systemic risk assessment frameworks early in AI system design. This work provides both empirical evidence and methodological infrastructure to advance robust, human-aligned AI risk governance.

Technology Category

• Humans and AI: Interaction Techniques and Devices
• Philosophy and Ethics of AI: Bias, Fairness & Equity
• Reasoning under Uncertainty: Other Foundations of Reasoning under Uncertainty

Application Category

• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

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.

Problem

• Analyzing gaps in AI risk reporting between developers and researchers
• Identifying overlooked real-world harms like fraud and manipulation
• Proposing structured risk reporting for human-interaction and systemic risks

Innovation

• Analyzed 460,000 AI model cards
• Built AI Model Risk Catalog
• Compared developer and researcher risks