Peer review, the volunteer backbone of academic publishing for the past 50 years, is straining as paper volume grows 5.6% per year in Scopus and Web of Science and AI-assisted writing accelerates submissions further. Researchers globally now donate roughly 15,000 years of labor to peer review annually, work valued at $1.5 billion for the US share alone. Editors say the system is not keeping up.
Steven Mack, an editor at Human Immunology, recently had to email 30 researchers to secure a single reviewer for one manuscript. Five years ago, he estimates 5 to 10 emails would land three willing reviewers. Mack, an immunogeneticist at the University of California, San Francisco, now receives a review request every day or two on top of his research role.
The volume problem is not confined to biology. Sebastian Lourido, a microbiologist at the Whitehead Institute, said he gets about 10 review requests per month and realistically has time for one or two. He described the current process as "extremely protracted and painful." Health economist Jason Semprini of Des Moines University had a paper on HPV vaccine mandates rejected after a single reviewer misread its central question — a failure mode that becomes more likely when editors cannot recruit two or three qualified reviewers per submission.
“I struggle like anything to get peer reviewers”— Haseeb Irfanullah, Editorial board member, Wiley's Learned Publishing
Key facts
- 01Papers indexed in Scopus and Web of Science are growing 5.6% annually, with global peer review consuming 15,000 years of labor worth $1.5 billion in the US alone.
- 02Human Immunology editor Steven Mack now emails 30 researchers to find one reviewer, versus 5 to 10 emails for three reviewers five years ago.
- 03Submissions to top AI conferences have risen 2- to 10-fold since 2019, overwhelming reviewer pools with cross-disciplinary work.
- 04Whitehead Institute microbiologist Sebastian Lourido receives roughly 10 review requests per month but has time for only one or two.
- 05The AI Alignment Forum, run by Lightcone Infrastructure, replaces peer review with an upvote-downvote-comment model.
Peer review's dominance is newer than its reputation suggests. Historian Aileen Fyfe of the University of St Andrews notes that scientific societies reviewed submissions as early as the 1800s, but the modern universal version only took hold in the 1970s. When the National Science Foundation faced political pressure over its spending — after post-World War II federal science funding had grown 25-fold — its longstanding practice of using external reviewers, dating to the 1950s, gave its grantmaking an air of legitimacy that kept lawmakers at bay.
Melinda Baldwin, a historian at the University of Maryland, framed the shift bluntly: peer review became science's brand promise, a way scientists sold their credibility to funders and the public. The United Kingdom followed a similar path in the late 1980s and early 1990s as controversies around HIV, autism, and cold fusion pushed institutions to draw a public line between reviewed and non-reviewed work.
That line is now smudging under the weight of output. Journals proliferated between 1960 and 2020, special issues generate custom demand, online-only publishing removes page constraints, and interdisciplinary research requires more expertise per review. AI compounds the pressure on both sides — it lowers the cost of writing and translating papers, and it enables paper mills that sell authorship on fabricated but plausible manuscripts. Haseeb Irfanullah, on the editorial board of Wiley's Learned Publishing, argues the industry may need to talk about "de-growth of publishing."
“just don't understand the field enough to evaluate, or give good feedback”— Haewon Jeong, Computer scientist, University of California, Santa Barbara
AI research is where the strain is most visible and where the workaround is furthest along. Submissions to top AI conferences have risen between 2- and 10-fold since 2019, and Haewon Jeong, a computer scientist at the University of California, Santa Barbara, said reviews increasingly come from people outside the subfield. Because much AI work happens in industry rather than academia, researchers are less dependent on peer-reviewed venues for career progression.
The exit path is the blog. Helen Qu, an AI researcher at the Flatiron Institute working on game-theoretic approaches to AI subversion, has decided to publish her research only on her personal blog. The AI Alignment Forum, run by Oliver Habryka's Lightcone Infrastructure, hosts a large community of academics and independent researchers writing on AI safety, using upvotes, downvotes, and comments in place of anonymous review.
Habryka acknowledged the model's obvious weakness: there is no guarantee a voter has read carefully, or read at all. Upvote-driven forums are vulnerable to popularity effects, in-group dynamics, and the same reviewer-fatigue problem in a different guise — engaged readers are a scarce resource everywhere. The traditional journal system, for all its bottlenecks, at least tries to require substantive written feedback.
Semprini, whose HPV mandate paper was killed by a single distracted reviewer, still calls peer review "the bedrock of science." His follow-up: "it's not standing strong." Reformers are floating everything from paid reviewing to structured AI-assisted first-pass triage to formally recognized post-publication review.
For AI companies, the shift matters commercially, not just academically. Frontier labs increasingly publish core results on their own blogs and arXiv rather than in peer-reviewed venues, and the field's most-cited safety and capability discussions now happen on forums and Substacks. If the informal channels harden into the primary record, journals lose their gatekeeping role in AI first — and the incentives around benchmarks, replication, and negative results move to whoever runs the largest audience, not the most rigorous review. That is a structural change in how AI knowledge gets validated, and it is already underway.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
