Big new paper release of Google for external agentic verification for science.
Rohan Paul Twitter · Rohan Paul (@rohanpaul_ai) · 2026-06-29
Google researchers publish a paper proposing agentic AI verification for scientific papers, introducing the Paper Assistant Tool, which found more known proof errors than single-model calls in pilots at STOC and ICML.
Extraction
Topics: agentic-verificationai-peer-reviewscientific-integrityllm-agentsai-science-tools
Claims
- AI is generating research papers faster than humans can review them, creating a 'verification debt' of unchecked claims and proofs.
- Google's proposed agentic verification framework splits papers into components, checks difficult sections deeply, and combines findings into a review.
- The Paper Assistant Tool focuses on objective checks such as proof errors, experimental gaps, and missing comparisons rather than accept/reject decisions.
- In tests on known errors, the Paper Assistant Tool found far more proof mistakes than a single model call could identify.
- Authors in STOC and ICML pre-submission pilots reported using the tool to fix serious theory gaps and run new experiments.
Key quotes
Science now needs AI review agents because AI is making papers faster than humans can check them.
The big deal is that scientific review may need its own AI stack, with review agents, clear roles, and human oversight, because paper generation is becoming partly automated too.
Paper Assistant Tool found far more known proof errors than a single model call, and many authors said it led them to fix serious theory gaps or run new experiments.