Updated
Updated · Google Research · Oct 14
Google Launches Deep Review Worldwide After 16,000-Paper Pilot as AI Submissions Top 74,000
Updated
Updated · Google Research · Oct 14

Google Launches Deep Review Worldwide After 16,000-Paper Pilot as AI Submissions Top 74,000

3 articles · Updated · Google Research · Oct 14

Summary

  • Google Research publicly released Deep Review, an agentic AI system that verifies complex documents including scientific papers, legal agreements, finance materials, Google Docs and presentations.
  • 16,000-plus manuscripts at STOC, ICML, NeurIPS and ICLR tested the tool under its earlier Paper Assistant Tool name, giving authors pre-submission feedback separate from official peer review.
  • 74,000 submissions hit ICLR, NeurIPS and ICML in 2026, up from about 45,000 in 2025, highlighting the validation bottleneck Deep Review is meant to ease as AI speeds document creation.
  • Deep Review uses a four-stage pipeline to check logical consistency, mathematical claims, factual grounding and comparisons with prior work, while Google says higher-level judgment remains with humans.
  • Gemini 4 Argon support is already boosting the system’s analysis, and Google said multi-file review across related documents, spreadsheets and exhibits is coming soon.

Insights

Can Google's new agentic AI truly solve the massive document review bottleneck that generative AI itself helped create across global industries?
With highly sensitive legal files now involved, how can Deep Review guarantee absolute data privacy without risking catastrophic multi-agent leaks?