Stanford Professor to Scale AI-Assisted Peer Review Starting December
Stanford professor Anshul Kundaje announced that starting in December he will use AI assistance to write as many peer reviews as possible in his subfield, beginning with 5-6 per month and potentially doubling as efficiency improves. He reports that 2-3 iterations with a frontier model yield finalized reviews that he fully vets himself, combining the strengths of human and AI. The effort could offer first-hand evidence on scaling AI into peer review.
Confirmed
- The plan starts in December with 5-6 reviews per month, possibly doubling later, limited to his subfield
- Workflow: 2-3 iterations with a frontier model, with all output personally vetted
- A concrete value point: the AI frequently catches inconsistencies between code and method descriptions
- He chose Substack as the platform, with no money involved; he declined a peer's suggestion to turn it into a funded journal, keeping it free and informal
- He plans to organize roughly 15-20 PIs in this collaborative AI review movement, with no central control over who reviews what, purely voluntary participation
- He believes the movement can cover most of the (subjectively) interesting papers in his subfield and become habitual over time
Why it matters
- This is a prominent scholar openly bringing frontier models into peer review at scale, providing a first-hand reference for feasibility and norms of AI-assisted academic review
- If the voluntary 15-20 PI collaboration works, it could seed open review practices outside the traditional journal system
2026-09-21 ~ 2026-09-21 · 5 related posts
Primary sources
- Stanford professor plans AI-assisted peer reviews of 5-6 papers monthly — anshulkundaje ·
- Stanford professor co-writes peer reviews with frontier models in 2-3 iterations, plans 15-20 PI network — anshulkundaje ·
- Researcher to publish AI-assisted peer reviews from December, starting at 5-6 papers per month on Substack — anshulkundaje ·
- [source] Stanford professor plans AI-assisted peer reviews of 5-6 papers monthly — anshulkundaje · 2026-09-21
- Professor: 2-3 iterations with frontier models yields high-quality peer reviews — anshulkundaje · 2026-09-21
- [source] Stanford professor co-writes peer reviews with frontier models in 2-3 iterations, plans 15-20 PI network — anshulkundaje · 2026-09-21
- Anshul Kundaje: community AI-assisted peer reviews to be run purely out of interest by 15-20 PIs — anshulkundaje · 2026-09-21
- [source] Researcher to publish AI-assisted peer reviews from December, starting at 5-6 papers per month on Substack — anshulkundaje · 2026-09-21