LLM-Based Analysis of the AI Incident Database: Insights for AI Governance
Bibliographic record
Abstract
Artificial Intelligence (AI) is increasingly adopted in critical sectors such as healthcare, finance, and public administration, where it promises significant gains in efficiency, automation, and decision support. At the same time, these systems expose societies to serious risks, including bias, discrimination, safety failures, and privacy infringements. To document and learn from such failures, the Artificial Intelligence Incident Database (AIID) was created as a community-driven repository and collective memory of AI harms, designed to support research, best practices, and governance. As of 2025, the AIID contains more than 1,100 incidents, yet its unstructured, narrative reports make systematic analysis difficult and limit their policy value. This paper addresses that challenge by applying a Large Language Model (LLM) pipeline, guided by the OECD AI Incident Reporting Framework, to transform AIID reports into structured data and enable systematic analysis of recurring patterns in AI incidents. The analysis reveals a sharp rise in frequency and severity since 2020, with human and economic harms dominating, transparency and fairness most frequently violated, and ICT, finance, and public administration accounting for most cases. In terms of implications, the study provides insights to better inform AI governance. It also demonstrates how to use LLMs to transform unstructured dataset into structured data for analysis.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.019 | 0.082 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.008 | 0.007 |
| Science and technology studies | 0.001 | 0.002 |
| Scholarly communication | 0.007 | 0.006 |
| Open science | 0.002 | 0.004 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.003 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".