Generative AI use and disclosure

A Lancet Correspondence auditing 2.5 million PubMed Central papers found fabricated citations rose from 1-in-2,828 papers in 2023 to 1-in-277 in early 2026 — and over 98% of flagged papers have seen no publisher correction.

A Nov. 2025 OpenReview bug exposed ICLR 2026 reviewer identities, enabling bribery and harassment amid separate reports that 21% of reviews were AI-generated.

The Supreme Court declined to hear Thaler v. Perlmutter (March 2026). What the human-authorship rule now means for CRediT, ICMJE, and AI disclosure practice.

In May 2026, arXiv began enforcing a one-year submission ban for authors whose papers show unchecked AI-generated content — hallucinated references or leftover chatbot text. Here’s what triggers it, how it’s enforced, and what it means for author accountability.

Agents4Science 2025, organized by researchers at Stanford and Together AI, ran as a one-day virtual conference on October 22, 2025 with an unusual submission rule: an AI system had to be the paper’s primary author, generating the hypotheses, running the experiments, and writing the manuscript, with a human listed as a supervising co-author. It is one of the most concrete tests to date of how authorship credit and disclosure norms hold up when the ‘researcher’ doing the work is an AI agent.

Zochi (Intology) and Sakana’s AI Scientist each passed real peer review in 2025 — one at ACL, one at an ICLR workshop. What it means for authorship.

Between late 2025 and early 2026, arXiv tightened endorsement (no more institutional-email-only qualification), required full English-language versions of every submission, and began requiring documented prior peer review for CS literature surveys and position papers after a flood of AI-generated submissions.

University of Melbourne has opened a formal probe into education researcher John Hattie over alleged plagiarism and fabricated AI-style citations.

ICML 2026 used hidden PDF watermarks to catch 506 reviewers using AI, desk-rejecting 497 papers tied to 398 of them amid a record 23,918 submissions.

ICML 2026 study: AI peer-review scores can be gamed by cosmetic ‘paper laundering’ rewrites, a risk authors warn could disadvantage novel work.

UHERO withdrew a Hawaii electricity-policy report in July 2026 after its author found an AI assistant had inverted a key solar-pricing trend from a Berkeley Lab dataset, unverified before publication — a live case study in AI-disclosure and verification gaps.

A new Learned Publishing study finds 83% of high-impact journals have explicit AI peer-review policies, versus 75% of mid-tier titles.

The ARC’s updated Generative AI policy took effect 28 April 2026, setting rules for applicants, assessors, and research offices.

Hachette, Cengage, Elsevier, and Scott Turow sued Google in SDNY, alleging Gemini trained on Google Books, Play Books, and Scholar without authorization.

COPE relaunched its paper mills working group in June 2026, citing AI-exploited coordinated misconduct as the driver, with a companion toolkit to its systematic-manipulation guidelines as the group’s first stated deliverable.

NIH’s May 2026 notice explains when AI use in funded research crosses into research misconduct, and what enforcement follows.

Wiley disclosed $49M in FY2026 AI licensing revenue ($110M lifetime). Taylor & Francis and Springer Nature show similar deals; small publishers see little.

A PNAS study of 5.2M papers found 70% of journals have AI policies, but only ~0.1% of papers actually disclosed AI use.

ICMJE updated its Recommendations in January 2026, adding a new standalone Section V on AI use and a new Section III.L.2 on authors’ access to trial data, plus revisions to sponsor-agreement, misconduct, trial-registration, and funding-disclosure guidance.

NeurIPS 2026 desk-rejected 178 position papers (18.4%) via the Pangram AI detector, no appeal allowed. What is verified, what is disputed, and why it matters.

A New York court reversed a student’s expulsion over a Turnitin false positive, and universities are switching AI-writing detectors off. The 2026 landscape.

Five major journals’ GenAI disclosure wordings, side by side. What each requires, where they differ, and the practical implications for authors.

When AI generated parts of a paper, which CRediT roles apply to the humans who prompted and verified? A working line between disclosure and contribution.

ICMJE rejects AI as an author; publishers require structured disclosure. A 2026 map of who asks for what, the hallucination problem, RAG, and watermark detection.