Topics
AI incidents by topic
234 documented incidents, grouped four ways. Every count is worked out from the library itself, so it is always current. See also the findings andthe open data.
Human behaviour
The human conduct the public record shows behind each incident, read against Headlights' fixed list of thirteen behaviours.
- Overconfidence 58 + 16
- Rules-behaviour gap 38 + 20
- Automation bias 28 + 5
- Normalisation of deviance 9 + 22
- Concealment 10 + 17
- Delayed intervention 8 + 16
- Defensive denial 5 + 16
- Diffusion of responsibility 5 + 13
- Judgement surrendered 12 + 5
- Incentive pressure 4 + 9
- Group conformity 4 + 4
- Authority ambiguity 5 + 1
- Confirmation bias 0 + 3
Industry
Where the AI system was used.
- Technology & software 38
- Consumer AI 36
- Government & public administration 32
- Retail, hospitality & travel 16
- Policing, justice & courts 16
- Professional & legal services 15
- Media & platforms 14
- Financial services & insurance 14
- Health & medicine 14
- Industry, transport & logistics 12
- Education 12
- Employment & hiring 12
- Housing & property 3
Governance domain
The governance question each incident raises.
- Algorithmic accountability 104
- Human oversight 96
- Consumer protection 59
- Data protection 44
- Civil liberties 37
- Identity & authenticity 36
- Decision authority 30
- Discrimination & profiling 28
- Administrative law 27
- Safety-critical systems 19
- Online safety 16
- Legal authority 12
- Marketing alignment 12
- Pricing governance 5
Country
Where the incident happened. Countries with two or more entries.