> ## Content Index
> Fetch the complete content index at: https://www.aigl.blog/llms.txt
> Use this file to discover other available public pages before exploring further.

# The AI Model Risk Catalog: What Developers and Researchers Miss About Real-World AI Harms
- URL: https://www.aigl.blog/the-ai-model-risk-catalog-what-developers-and-researchers-miss-about-real-world-ai-harms/
- Published: 2025-09-22T11:13:35.000Z
- Updated: 2026-09-12T20:06:00.000Z
- Description: An analysis of nearly 460,000 AI model cards shows developers overwhelmingly report technical risks—while real-world harms arise more from misuse and misinformation.
- Author: Jakub Szarmach
- Tags: Research, #aigl-library

[The AI Model Risk Catalog- What Developers and Researchers Miss About Real-World AI HarmsThe AI Model Risk Catalog- What Developers and Researchers Miss About Real-World AI Harms.pdf1 MBdownload-circle](https://www.aigl.blog/content/files/2025/09/The-AI-Model-Risk-Catalog--What-Developers-and-Researchers-Miss-About-Real-World-AI-Harms.pdf "Download")

## ⚡ **Quick Summary**

This paper introduces the *AI Model Risk Catalog*, the largest dataset of developer-identified AI risks to date, derived from over 64,000 Hugging Face model cards. The authors compare these developer-reported risks to two other key sources: the *MIT Risk Repository* (capturing researcher-envisioned risks) and the *AI Incident Database* (cataloging real-world harms). The study reveals that while developers focus on technical and bias-related risks, both developers and researchers underreport harms stemming from human interaction—particularly fraud, misinformation, and malicious use—despite their prevalence in real incidents. This gap underscores a need for structured, contextualized risk reporting that accounts for real-world usage dynamics.

## 🧩 **What’s Covered**

The study builds a new AI Model Risk Catalog by:

1. **Extracting Risks**: Analyzing 64,116 model cards from Hugging Face (2024 snapshot) with risk-related sections, and isolating 2,863 unique risk mentions after de-duplication and validation.
2. **Taxonomizing Risks**: Using both the *DeepMind taxonomy* (Weidinger et al.) and the *MIT AI Risk Repository taxonomy* to categorize risks. Most fall under “Discrimination & Toxicity” (44%) or “AI System Safety, Failures & Limitations” (37%) .
3. **Comparative Analysis**: Contrasting the catalog with:This reveals divergences in focus:
  - The *MIT Risk Repository* (967 research-defined risks)
  - The *AI Incident Database* (869 real-world cases)
  - **Developers** focus on technical faults and bias.
  - **Researchers** prioritize systemic and ethical concerns.
  - **Real-world harms** stem mostly from **misinformation** and **malicious use**—areas underrepresented in both expert groups .
4. **Risk Reporting Trends (2022 → 2024)**:
  - Risk reporting quality has **declined**—fewer unique entries and shorter descriptions.
  - Developers increasingly reuse generic risk templates, especially for fine-tuned LLMs.
  - The share of risk categories has shifted: bias risks increased, while misinformation risks dropped by 3x .
5. **Modality-Specific Risks** (Page 15):
  - Non-text modalities (e.g., image, audio, video) are increasingly associated with risks like **privacy**, **security**, and **malicious use**.
  - Multimodal models saw a 3x increase and are more likely to raise concerns about **fraud** and **identity manipulation**.
6. **Recommendations**:
  - Urge developers to adopt **structured** and **contextualized** risk descriptions.
  - Encourage researchers to focus more on **social engineering** and **human interaction risks**.
  - Propose integration with auditing tools like *RiskRAG*, *ExploreGen*, and *BenchmarkCards*.

## 💡 **Why it matters?**

The AI safety ecosystem remains siloed: developers, researchers, and incident reporters speak different “languages” of risk. This paper’s comparative framing exposes a critical gap—**real-world AI harms are dominated by misuse and manipulation**, yet developers and researchers focus on system-centric or ethical concerns. Without bridging this gap, risk governance frameworks (like the EU AI Act or NIST RMF) may remain blind to the most urgent and prevalent dangers. The *AI Model Risk Catalog* provides an urgently needed resource to **ground risk analysis in actual model behavior** and real-world impact, offering a template for more actionable, interdisciplinary AI assurance practices.

## ❓ **What’s Missing**

- **User and community perspectives** are absent—no input from civil society, NGOs, or end-users.
- **Risk severity and likelihood** are not quantified—hindering prioritization.
- **Linkage to regulatory frameworks** like ISO 42001 or the EU AI Act is not directly explored.
- The catalog excludes proprietary or enterprise models, which may present **higher-stakes risks**.
- The study doesn’t test how well cataloged risks help in **predicting or mitigating future incidents**.

## 👥 **Best For**

- **AI developers** seeking to improve model documentation and risk foresight
- **Researchers** exploring the divergence between theoretical risk and real-world impact
- **Auditors and regulators** aiming to align assurance practices with actual harms
- **Journalists and analysts** investigating emerging gaps in AI governance
- **Toolmakers** designing risk assessment and LLM red-teaming platforms

## 📄 **Source Details**

- **Title**: The AI Model Risk Catalog: What Developers and Researchers Miss About Real-World AI Harms
- **Authors**: Pooja S. B. Rao, Sanja Šćepanović, Dinesh Babu Jayagopi, Mauro Cherubini, Daniele Quercia
- **Institutions**: University of Lausanne, Nokia Bell Labs, IIIT Bangalore, University of Oxford, Politecnico di Torino
- **Date**: August 21, 2025
- **Availability**: [https://social-dynamics.net/ai-risks/catalog](https://social-dynamics.net/ai-risks/catalog?ref=aigl.blog)
- **Preprint**: arXiv:2508.16672v1

## 📝 **Thanks to**

Pooja S. B. Rao and the team at Social Dynamics Lab for an indispensable tool in evidence-based AI risk reporting.