Categories
Incidents in the Index are classified into ten primary categories that reflect the nature of the harm or failure documented.
Discrimination and fairness
Unfair or unequal outcomes for individuals or groups, including misrepresentation and unequal performance across groups.
140 incidents
Harmful and abusive content
Toxic, sexual, abusive or exploitative material produced or spread by an AI system, including non-consensual imagery and harms to children.
927 incidents
Privacy and surveillance
Personal or confidential data exposed, inferred or collected without consent, and surveillance uses such as facial recognition.
839 incidents
Security and cyberattacks
Vulnerabilities in AI systems, prompt injection and jailbreaks, and AI used to enable attacks or weapons.
1208 incidents
Misinformation and synthetic media
False or misleading content, deepfakes and influence operations that degrade the information space.
1221 incidents
Fraud, scams and manipulation
Voice cloning, impersonation, targeted scams and social engineering carried out with AI.
1079 incidents
Reliability and hallucination
Fabricated facts and citations, wrong outputs, and failures of robustness or transparency that caused harm.
486 incidents
Safety and wellbeing
Physical injury or death, psychological harm, overreliance, loss of agency and loss of control involving AI systems.
1012 incidents
Intellectual property and appropriation
Copyright and training-data disputes, likeness and voice rights, and the devaluation of human work.
417 incidents
Governance, accountability and labour
Oversight failures, lack of transparency from deployers, employment effects, market concentration and environmental cost.
341 incidents