ICLR · ICML · NEURIPS — 23 CONFERENCE‑YEARS, CLASSIFIED
0%
of the 55,794 papers accepted at ICLR, ICML and NeurIPS, 2019–2026, are AI safety — 2,328 papers.
Safety's share of the field has grown roughly 28-fold since 2019. A language model reads every accepted paper and sorts it into one of four categories; the safety papers are then divided into seventeen research areas.
The share of accepted papers focused on AI safety, across ICLR, ICML & NeurIPS · 2019–2026.