AI hallucination and reliability failures
Definition
Hallucination incidents document cases where an AI system produced outputs that were factually incorrect, fabricated, or misleading in ways that caused measurable harm, material misrepresentation, or erosion of trust in a consequential context.
Included
- False citations or fabricated legal, medical, or scientific references that were acted upon
- Incorrect factual claims generated by AI that influenced decisions with real-world consequences
- AI-generated content that misrepresented events, statistics, or named individuals
- Failures in AI-powered navigation, diagnosis, or recommendation systems due to incorrect outputs
- AI agents that confidently executed incorrect actions based on hallucinated facts
Excluded
- Minor inaccuracies in low-stakes contexts with no material consequence
- Outputs that were ambiguous or incomplete but not factually false
- Failures of intent, tone, or style rather than factual accuracy
Incident records
Category-level filtering is not yet available in this dataset release. Showing the 5 most recent incidents across the full index.