No. 006
AAAI-26 AI review pilot, ScienceOne 100, NEJM retraction
Peer Review
-
AI-Assisted Peer Review at Scale: The AAAI-26 AI Review Pilot
arXiv, April 15 2026
The first large-scale field deployment of AI peer review processed all 22,977 AAAI-26 main-track submissions in under a day, and a post-conference survey found participants rated the AI reviews higher than human reviews on technical accuracy and research suggestions.
-
Illicit Use of AI by Philosophers Refereeing for Journals
Daily Nous, April 29 2026
A philosophy PhD student discovered their journal review was likely AI-generated, with detection tools rating it 100% AI-written, a problem previously documented in computer science that appears to have spread to the humanities.
-
You can just review things: A digital ethnography of informal peer review
arXiv, April 2026
Patel and Chan's cross-platform ethnography across 15 communities documents how scholar-organized informal peer review actually happens on Bluesky, Reddit, preprint comments, and personal blogs. Their empirical anchor: PubPeer has comments on only 0.33% of indexed Scopus papers, while Bluesky has 58× Twitter's research-mention rate. A qualitative companion to formal accounts of peer review's rate problem.
Tools & Infrastructure
-
Chinese Academy of Sciences unveils AI model system for scientific research
Guangming Online, April 29 2026
CAS deployed ScienceOne 100, a research AI platform spanning eight scientific domains with over 2,000 tools across 50+ institutes, making it one of the largest coordinated efforts to embed AI into a national research system.
-
arXiv, April 29 2026
An open-source toolkit for detecting fabricated citations in AI-assisted research runs on a laptop CPU in seconds, released under Apache 2.0. A practical response to the growing problem of LLM-hallucinated references in scholarly writing.
Publishing & Integrity
-
NEJM retracts case study for AI-manipulated imagery
Retraction Watch, May 1 2026
The New England Journal of Medicine retracted a clinical image case study after authors admitted using AI to reposition a measuring tape in a photograph, the journal's first retraction since 2020, flagged by a PubPeer commenter who noticed irregularities in the ruler's numbering.
-
Rethinking Publication: A Certification Framework for AI-Enabled Research
arXiv, April 23 2026
A proposed two-layer framework separates knowledge quality assessment from grading of human contribution, categorizing AI-enabled research into three tiers based on the degree of human involvement: an attempt to give publication systems a vocabulary for work that is valid but not entirely human-authored.