Who pays when the AI is wrong?
AI regulation, trustworthy AI, responsible AI and the EU AI Act appear on 229 recent papers, mostly in AI and management journals, but only 12 of them carry an insurance, liability, compensation or damages keyword, and AI liability on 2. Rules now assign duties; they say little about who pays for harm. Can AI harms be insured, and would insurance discipline deployment?
What prompted this
Since October 2025, 229 papers carry an AI-regulation, AI-safety, EU AI Act, AI-alignment, responsible-AI or trustworthy-AI keyword (157 in AI journals, 53 in management): "AI regulation" on 53, "trustworthy AI" on 51, "responsible AI" on 46, "AI safety" on 28, "EU AI Act" on 25. Only 12 of the 229 also carry an insurance, liability, tort, compensation, legal-responsibility or damages keyword, and "AI liability" appears on 2 papers.
Questions
- For harms already litigated (wrong credit decisions, biased hiring tools, chatbot misinformation, autonomous vehicle crashes), who paid, and how long did it take?
- Could insurers price AI liability given correlated failures across customers of the same model, and would premiums vary with testing and monitoring practices?
- Does the withdrawal of the EU's AI liability directive leave victims with workable claims under existing law?
Data that could help
Court records and settlements involving AI systems, insurance policy wordings and exclusions, and the texts of liability rules across jurisdictions.
Next step
Compile a database of litigated AI harms since 2018 with the outcome, the party that paid and the time to resolution.