Research map / Large language models / AI and society
Lessons for Editors of AI Incidents from the AI Incident Database
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Reviews the AI Incident Database's 750+ incidents and its CSET and GMF taxonomies to identify editorial challenges (temporal ambiguity, multiplicity, societal harms, epistemic uncertainty) and proposes lessons for AI incident reporting.
Problem
Defining, indexing, and analyzing AI incidents is challenging: harm events have ambiguous timing, multiplicity, aggregate societal harms, and incomplete information, and no widely shared incident reporting practice exists as AI incident reporting becomes legislated.
Approach
Practitioner reflection grounded in the AIID's incident dataset and two taxonomies: it groups incidents that stress the editing methodology into four qualitative challenge themes and examines how the CSET and GMF taxonomies handle or are challenged by them, deriving lessons for future reporting practice.
Key results
- Identifies four recurring editorial challenges in AIID incidents: temporal ambiguities, multiplicity, aggregate and societal harms, and epistemic uncertainty.
- Reports that the CSET AI Harm Taxonomy covers 214 AIID incidents and the GMF taxonomy covers 188 incidents at time of writing.
- Notes that the GMF taxonomy explicitly marks annotations as 'potential' or 'known' to address epistemic uncertainty.
Surprising findings
- Inter-annotator variation can be reduced but not removed even with iteratively developed taxonomy definitions and guidelines.
Limitations
- Qualitative themes are tentative and not systematically validated across incidents
- Findings are drawn from the authors' own AIID editorial experience, which may limit generality
Experimental setup
Datasets: AI Incident Database (750+ incidents)
Builds on: CSET AI Harm Taxonomy Goals, Methods, and Failures taxonomy CVE Program
Closest papers by meaning
- When Your AI Becomes a Target: AI Security Incidents and Best Practices (AAAI 2024 · similarity 0.86)
- Evaluation and Incident Prevention in an Enterprise AI Assistant (AAAI 2025 · similarity 0.84)
- SafetyAnalyst: Interpretable, Transparent, and Steerable Safety Moderation for AI Behavior (ICML 2025 · similarity 0.81)
- Merging AI Incidents Research with Political Misinformation Research: Introducing the Political Deepfakes Incidents Database (AAAI 2024 · similarity 0.81)
- Position: In-House Evaluation Is Not Enough. Towards Robust Third-Party Evaluation and Flaw Disclosure for General-Purpose AI (ICML 2025 · similarity 0.79)
- SLS-Bench: A Benchmark for Incident Log Summarization with Synthetic Observability Data (NeurIPS 2026 · similarity 0.79)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.