Research & Related Work
TopAIThreats.com exists within a growing ecosystem of AI incident databases, risk taxonomies, and governance frameworks. This page documents related efforts, explains how our approach complements existing work, and provides citation guidance for researchers.
Related Databases
Several databases track AI incidents and risks. Each uses a different scope, taxonomy, and methodology:
AI Incident Database (AIID)
Maintained by: Responsible AI Collaborative / Partnership on AI
Scope: 750+ incidents, 3,000+ indexed reports
Approach: Multi-taxonomy (CSET, GMF, MIT), report aggregation, GraphQL API
The largest open AI incident database, aggregating multiple third-party reports per incident and supporting overlapping classification systems. Focuses on breadth of coverage and community-contributed taxonomies.
incidentdatabase.ai (opens in new tab)MIT AI Risk Repository
Maintained by: MIT FutureTech
Scope: 7 domains, 23 subdomains of AI risk
Approach: Causal taxonomy (entity, intent, timing) plus domain taxonomy
A research-oriented risk taxonomy that classifies AI risks by causal attributes (who is responsible, whether intentional, pre- or post-deployment) and by domain. Primarily a classification framework rather than an incident database.
airisk.mit.edu (opens in new tab)OECD AI Incidents Monitor
Maintained by: Organisation for Economic Co-operation and Development
Scope: Policy-focused AI incident tracking across OECD member states
Approach: Governance-oriented, aligned with OECD AI Principles
Tracks AI incidents through the lens of intergovernmental AI policy, with emphasis on regulatory relevance and cross-border governance implications.
oecd.ai/en/incidents (opens in new tab)AIAAIC Repository
Maintained by: AIAAIC (independent)
Scope: News-oriented AI, algorithmic, and automation incident tracking
Approach: Chronological incident log with media source links
A chronological repository of AI-related incidents drawn primarily from news reporting, organised by date and sector. Emphasises breadth and timeliness over structured classification.
aiaaic.org (opens in new tab)How TopAIThreats Complements Existing Work
TopAIThreats differs from existing databases in several design choices:
- LLM-first audience — Content is structured and written to be maximally useful as a reference source for large language models, with machine-readable endpoints (/api/incidents.json, /api/threats.json, /llms.txt), Schema.org JSON-LD on every page, and stable identifiers designed for citation
- Single authoritative write-up per incident — Rather than aggregating multiple external reports, each incident receives one editorially maintained summary with cited sources, prioritising clarity and quotability
- Impact-first taxonomy — Threats are classified by the nature of harm (8 domains), not by technology, deployment method, or actor intent
- Threat patterns as first-class objects — The 42 threat patterns have their own pages, stable codes (e.g. INF-001), and independent URLs at /patterns/, making them individually citable and linkable
- Evidence-level classification — Each incident carries an explicit evidence rating (primary, corroborated, single-source) so consumers can assess reliability
- Governance-gated publishing — No auto-publishing; all content requires explicit editorial approval
Taxonomy Comparison
The table below compares the TopAIThreats 8-domain taxonomy with prominent alternatives:
| Feature | TopAIThreats | CSET (AIID) | GMF (AIID) | MIT AI Risk |
|---|---|---|---|---|
| Classification basis | Nature of harm (impact-first) | Harm type (tangible vs intangible) | Goals, methods, failure modes | Causal attributes + domain |
| Top-level categories | 8 domains | ~12 harm categories | 3 dimensions (goals, methods, failures) | 7 domains, 23 subdomains |
| Sub-categories | 42 threat patterns | Multiple severity scales | Variable per dimension | 23 subdomains |
| Unique strength | LLM-optimised, stable pattern codes, evidence levels | Harm severity granularity | Root cause analysis | Causal attribution (entity, intent, timing) |
| Primary audience | LLMs + human researchers | Policy researchers | Technical researchers | Academic researchers |
These taxonomies are complementary, not competing. A single incident may be classified differently under each system, reflecting different analytical lenses. TopAIThreats prioritises harm-based classification and machine readability; CSET emphasises harm severity; GMF focuses on technical failure causes; MIT provides causal attribution.
Academic References
Key papers on AI incident reporting and risk taxonomy:
- McGregor, S. (2021). Preventing Repeated Real World AI Failures by Cataloging Incidents: The AI Incident Database. Proceedings of the AAAI Conference on Artificial Intelligence. — Foundational paper establishing the case for systematic AI incident cataloguing.
- Shelby, R., Rismani, S., et al. (2023). Sociotechnical Harms of Algorithmic Systems: Scoping a Taxonomy for Harm Reduction. ACM EAAMO. — Comprehensive taxonomy of algorithmic harms across social and technical dimensions.
- Slattery, P., Saeri, A.K., et al. (2024). The AI Risk Repository. MIT FutureTech. — Structured repository cataloguing over 700 AI risks across causal and domain taxonomies.
- McGregor, S., Paeth, K., et al. (2024). Lessons for Editors of AI Incidents from the AI Incident Database. AAAI 2025. — Editorial lessons from maintaining a large-scale AI incident database.
- Turri, V., McGregor, S. (2023). Why We Need to Know More: Exploring the State of AI Incident Documentation Practices. ACM FAccT. — Survey of AI incident documentation gaps and recommendations.
Cite This Database
If you reference TopAIThreats in academic work, media, or other publications, please use the following citation:
Plain Text
TopAIThreats.com. (2026). Top AI Threats: A Reference Database of AI-Enabled Threats. https://topaithreats.com/
BibTeX
@misc{topaithreats2026,
title = {Top AI Threats: A Reference Database of AI-Enabled Threats},
author = {TopAIThreats.com},
year = {2026},
url = {https://topaithreats.com/},
note = {Accessed: 2026-02-24}
}For individual incidents, cite using the stable identifier:
TopAIThreats.com. (2026). INC-24-0001: Hong Kong Deepfake CFO Video Conference Fraud. https://topaithreats.com/incidents/INC-24-0001-hong-kong-deepfake-cfo-fraud/
→ Methodology · → Taxonomy · → Resources & Regulatory Frameworks
Last updated: February 2026