Google opens Deep Review, its agentic AI for verifying technical documents, to the public
Google Research has publicly launched Deep Review, an agentic system tested on more than 16,000 conference submissions, to check the logic, math and factual grounding of complex scientific, legal and financial documents.
Google Research on October 14, 2026 publicly released Deep Review, an agentic AI system designed to provide rigorous feedback and technical verification on complex documents.
From pilot to public release
Deep Review is the evolution of Google's experimental Paper Assistant Tool (PAT). Over the past year, Google partnered with four flagship computer science conferences — STOC 2026, ICML 2026, NeurIPS 2026 and ICLR 2027 — to give private, pre-submission feedback to authors of more than 16,000 manuscripts who opted in. Google said the feedback was fully separate from official peer review, and that PAT did not collect the papers for training.
Google frames the launch as a response to a "validation bottleneck": AI has accelerated content creation while human capacity for verification has stayed largely fixed. This year, the company noted, there were more than 74,000 submissions to ICLR, NeurIPS and ICML, up from about 45,000 in 2025.
How it works
Deep Review uses a four-stage agentic pipeline:
- Document segmentation: the document is decomposed into logical segments.
- Adaptive budgeting: compute is allocated by information density, so dense proofs get more reasoning than an introduction.
- Deep parallel verification: specialized agents review each segment in parallel with access to the full document context.
- Global synthesis and grounding: a synthesis agent consolidates findings, runs cross-section consistency checks and performs external grounding to guard against hallucinations.
Access and data policy
Every user gets two free document reviews per week, and high-volume users can connect a Gemini API key for higher limits and programmatic access through the Deep Review API. PDFs and plain text are supported at launch, with multi-file inputs coming soon. Google says uploaded documents are never used to train or improve its models and are stored for one year solely for display in the Deep Review interface.
Deep Review now supports Gemini 4 Argon, and Google says it can be applied beyond research papers to design documents, presentations, and legal and financial documents. The company also introduced a related multi-agent document review pattern in its Antigravity platform.
Source: Google Research.
Related articles
Mayhem 10 drones get upgraded to connect satellite, cellular and radio simultaneously
AeroVironment is preparing to strengthen communications on its Mayhem 10 autonomous military drone system with...
US optical chip gyroscope gets 2,000-fold sensitivity boost, detects slow rotation
Researchers at the University of California, Santa Barbara, and Caltech have developed a chip-scale optical...
Anthropic says new defenses cut distillation of Claude’s reasoning by 90% to 99%
Anthropic says safeguards rolled out in recent months have reduced the data harvested by distillation attackers by around 90% to 99%, naming Chinese labs Zhipu and Moonshot as examples of industrial-scale efforts to copy Claude’s capabilities.
Microsoft unveils Maia 300 AI chip, plans multi-gigawatt in-house deployments from 2027
Microsoft announced Azure Maia 300, the third generation of its custom AI chip, and said it will scale the Maia family to multi-gigawatt deployments starting in 2027, making it the primary inference system for its products and services.
MI5 launches AI Threat Assessment Centre as spy chief flags AI-enabled bioweapons risk
MI5 Director General Sir Ken McCallum used his annual threat update to announce the AI Threat Assessment Centre (AITAC), bringing together UK expertise to assess AI national-security risks, and said he gives the greatest weight to AI's potential to change the stakes on bioweapons.
Anthropic’s third-quarter revenue jumps 45% to $16.8 billion ahead of planned IPO
Anthropic told prospective investors at a pre-IPO meeting that third-quarter revenue rose 45% from the prior quarter to $16.8 billion and that it was profitable on an adjusted basis, Reuters reported.