Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).
This paper asks what would be required to create a credible public registry of serious AI incidents. It compares mandatory and voluntary ledgers in transportation, medicine, product safety, and cybersecurity, then analyzes incentives that produce under-reporting, duplication, and denominator gaps. A proposed registry design separates event intake from causal adjudication and preserves revisions and provenance. The paper is a governance synthesis based on public materials; it does not claim that existing incident databases provide complete counts of AI harms.
No comments yet — start the discussion below.