OPEN_SOURCE ↗
REDDIT · REDDIT// 35d agoRESEARCH PAPER
Journals' AI disclosure rules fall short
A new PNAS analysis of 5.2 million papers and 75,000 full texts found AI-assisted scientific writing has surged since 2023 while disclosures remain almost nonexistent. The paper argues journal mandates alone are not changing researcher behavior, pushing the debate from simple disclosure rules toward broader governance of generative AI in science.
// ANALYSIS
This is less a story about cheating than about policy lag: AI writing is already normal in research workflows, and journal rules are barely registering.
- –The striking number is the gap between detected usage and explicit disclosures: only 76 of 75,000 recent papers disclosed AI writing help.
- –Because the study found little difference between journals with and without disclosure policies, compliance-by-policy looks weak without enforcement or incentives.
- –Faster uptake in non-English-speaking contexts suggests generative AI is functioning as both a writing aid and a scientific access equalizer.
- –For developers building research tools, the opportunity is auditability: provenance, disclosure UX, and verifiable AI-use logs will matter more than generic text generation.
// TAGS
researchregulationethicsllmpnas-ai-disclosure-study
DISCOVERED
35d ago
2026-03-07
PUBLISHED
35d ago
2026-03-07
RELEVANCE
7/ 10
AUTHOR
monkey_spunk_