Stanford Professor Boycotts Journals That Ban AI

Episode 1 · Stanford Professor Refuses to Review for AI-Banning Journals, Sparking Academic Debate (2026-10-28, 20 posts)

Starting October 28, Stanford professor Anshul Kundaje posted a lengthy thread on X publicly announcing he will refuse to review for journals that ban AI use, and will soon stop submitting to them; he argues AI-assisted peer review is on average higher quality, and blanket bans are both unenforceable and backwards. The remarks triggered days of back-and-forth debate among scholars, becoming a concentrated clash over the boundaries of AI participation in academic publishing.

Confirmed

• Kundaje's core positions: refuse to review for journals banning AI; criticize publishers for extracting free labor from scholars while profiting from public funds; advocate allowing AI use with incentives for honest disclosure rather than prohibition.
• He acknowledged that delegating all thinking to AI from the start introduces bias and degrades thinking ability, recommending completing one's own thinking and writing first, then augmenting with AI; he believes human-AI collaborative review can outperform three human reviewers.
• He repeatedly stressed that review is not a one-way process—authors have rebuttal rights over both human and AI feedback; he also noted that overconfident AI usually concedes when rebutted, while overconfident humans do not.
• Pangram reported that 21% of ICLR 2026 reviews were fully AI-generated; an independent survey by Shen and Wang estimated around 20% (see original posts for exact figures). Bioinformatician Heng Li's opposition to journals fully banning AI-generated reviews also drew controversy.

Unconfirmed

• No consensus on average quality of AI vs. human reviews: bidumit believes AI reviews already beat the average human reviewer, especially for niche and interdisciplinary work; djhshih countered from experience that AI assistance makes good reviewers better but mediocre ones worse—the latter confidently pontificate on things they clearly don't understand and are harsher, often on interdisciplinary papers; Kundaje responded that this is 100% wrong.

Why it matters

• The dispute cuts to the heart of peer review: jmschreiber91 distinguished fully-AI reviews from AI-assisted ones, noting that accepting an invited review and dumping the paper into Claude is close to dishonest, and journals could offer automated review themselves; javidlab warned pure-AI review is narrower and more gameable, and handing review entirely to AI would drag academia into an innovation trough.
• hexane360 proposed a more radical governance approach: journals run AI review centrally and bar individual human reviewers from using AI; Kundaje called this flatly wrong—journals should not become monopolists of review discourse. With roughly 20% of ICLR reviews suspected AI-generated, "disclose or ban" has become a question academic publishing must answer.

• ~20% of ICLR reviews may be AI-generated: Heng Li pushes back, expert says AI review can beat average humans — anshulkundaje · 2026-10-28
• Stanford professor boycotts journals that ban AI, moves to open AI-assisted peer review — anshulkundaje · 2026-10-28
• AI-only peer review risks an 'innovation valley' that academia can't climb out of — javid_lab · 2026-10-28
• Stanford professor refuses to review for journals banning AI, slams publishers' 'highway robbery' — anshulkundaje · 2026-10-28
• Stanford prof pushes back on banning AI in peer review: lazy reviewers were always lazy — anshulkundaje · 2026-10-28
• AI makes good reviewers better but average ones worse, researcher observes — djhshih · 2026-10-29
• AI-assisted peer review debate: does it make mediocre reviewers worse? — anshulkundaje · 2026-10-29
• AI reviewers will concede when rebutted; overconfident human reviewers won't — anshulkundaje · 2026-10-29
• Debate over AI peer review: AI reviews may beat average human reviewers — jmschreiber91 · 2026-10-29
• The AI peer review debate: is just running a paper through Claude dishonest? — anshulkundaje · 2026-10-29
• Stanford prof: journals should disclose AI in peer review, not ban it — anshulkundaje · 2026-10-29
• Stanford professor: journals should allow AI in review with honest disclosure incentives — anshulkundaje · 2026-10-29
• AI reviews now beat the average human reviewer, especially in cross-disciplinary work — bidumit · 2026-10-29
• Stanford professor: peer review was never one-way — authors rebut AI reviews too — anshulkundaje · 2026-10-29
• Stanford prof: think first, then use AI—delegating thinking early erodes your mind — anshulkundaje · 2026-10-30
• Debate: journals should run AI review themselves and bar human reviewers — hexane360 · 2026-10-30
• Debate: Should Journals Run AI Reviews and Bar Human Reviewers From Using AI? — anshulkundaje · 2026-10-30
• Stanford Prof Anshul Kundaje: Banning AI in Peer Review Is Futile, Incentivize Disclosure — anshulkundaje · 2026-10-30
• Stanford Prof argues human+AI collaborative peer review beats three human reviewers — anshulkundaje · 2026-10-30
• Stanford prof Anshul Kundaje: most papers don't need formal review, the system is overwhelmed — anshulkundaje · 2026-10-30

Episode 2 · Stanford Researcher Sees Step-Change in GPT6 Astra's Peer Review Quality (2026-10-29, 2 posts)

Stanford professor Anshul Kundaje reports that GPT6 Astra (Ultra) marks a step-change in LLM paper review quality, and that multi-turn debate with the model yields far better reviews than a single output.

• Researcher: GPT6 Astra brings a step change in review quality, guided prompts help further — anshulkundaje · 2026-10-29
• Stanford prof's LLM peer-review workflow: argue back and forth, bucket issues by severity — anshulkundaje · 2026-10-29