The searchable record of real-world AI incidents.
Explore documented AI failures, security events, privacy incidents, harmful outcomes and governance breakdowns across companies, AI systems, industries and countries.
The complaint against AI-powered smart glasses is not that they record: they advertise that feature prominently. The complaint is that people nearby had no way to know whether recording was active at any given moment, an...
- Date
- 22 September 2026
- Source
- OECD
- Reviewed by
- Shiva Ganesh
Explore the Index
Discover documented incidents by category, organization, regulation and source.
Recently documented
What the data shows
Patterns across the incidents published so far. The dataset grows as research is completed, so these figures describe the Index, not the world.
| Tesla | 5 | |
| AI Now Institute | 2 | |
| OpenAI | 2 | |
| Amazon | 1 | |
| Anonymous extortionist | 1 | |
| Arkansas Supreme Court | 1 | |
| China's Ministry of Industry and Information Technology | 1 | |
| Cruise | 1 |
Based on incidents currently recorded in the AI Incident Index. This dataset does not represent the total universe of AI incidents.
Research from the Index
Evidence before conclusions.
Every incident is connected to public evidence and classified using a consistent methodology.
Counts describe what has been documented in this dataset, never an estimate of how often AI incidents occur.
- Documented
- Every published record is a fully written article, not a copied source blurb.
- Sourced
- Each record names its source database and links the origin document where one is public.
- Reviewed
- A named curator and reviewer stand behind every record.
- Classified
- Records carry an organization and, where the evidence supports one, an editorial category.
- Citable
- APA, MLA, BibTeX and plain-text citations on every incident page.
- Correctable
- Anyone, including the organizations named, can submit a correction at no cost.
Help improve the public record.
Know of an incident we should include, or found something that needs correcting?