Eight-body comparison · every row re-verified against primary sources 29 July 2026
Who requires what, when authors use AI in a manuscript
A side-by-side reference comparing AI-use disclosure requirements across two standards bodies (STM, currently a classification framework, not a mandate), two ethics/editorial organizations (COPE, ICMJE), three voluntary/emerging proposals (DAIU, ai-label.org, Earned Trust (AIast) — an independent researcher's role-based byline standard, not yet adopted by any journal, though publicly archived on Zenodo), and two major publishers (Elsevier, Springer Nature). Every cell links back to the primary source it was drawn from.
Reading the dashboard
Compliance matrix — use-case × body
Click any cell to filter the detailed table below to that exact use-case and body. Cells with multiple documented rows show a "+N" count.
Quick check: Elsevier vs. Springer Nature
Full requirement table
Sortable, filterable, and searchable. Every row cites the exact source it was drawn from.
| Body | Use-case category | Specific use-case | Requirement | Format | Mandatory fields | Placement | Source |
|---|
No requirements match these filters. Try clearing a filter or search term.
Body-by-body snapshot
A shared standard is coming — and Round 2 is open now
The International Science Council, COPE, STM and the Global Young Academy are jointly developing a Global Reporting Standard for AI Disclosure in Research — now widely referred to as the Vancouver Standard, after the World Conference on Research Integrity that hosted its focus track in Vancouver on 3–6 May 2026, across two sessions on disclosure thresholds and on accountability and verification.
The consultation runs in three rounds. Round 1 (December 2025 – February 2026) mapped what authors and publishers need disclosed, and is closed. Round 2 is open as of July 2026 and closes 16 October 2026, addressing what information should be disclosed and the content taxonomies to carry it. Round 3 follows at the end of 2026 into early 2027 for final refinement. Input is submitted through the ISC's webform, and the organisers ask respondents to coordinate at the organisational level where possible.
Until it lands, disclosure requirements remain fragmented across the bodies and publishers above — always check the specific journal's current author guidelines before submission.
Sources: International Science Council — AI disclosure in research · WCRI 2026 focus track · The Scholarly Kitchen, 13 July 2026 · STM, 27 January 2026
Primary sources
Every value in this dashboard was drawn directly from these documents. Verify against the live policy page before relying on it for submission.
Exactly which document version each row was read from
Checked 29 July 2026. Where a body publishes an update date, it is given here so a reader can tell at a glance whether this dashboard reflects their current guidance.
- Elsevier
- Generative AI policies for journals — page states “Policy updated June 2026.”
- Springer Nature / Nature Portfolio
- AI editorial policy — no published revision date; policy states it is reviewed on an ongoing basis.
- ICMJE
- Recommendations, January 2026 revision — Section V, Use of Artificial Intelligence in Publishing.
- COPE
- Position statement, Authorship and AI tools — version 1, published 20 August 2024 (last reviewed 13 February 2023). Still the current position.
- STM
- Recommendations for a Classification of AI Use in Academic Manuscript Preparation, 19 September 2025 — still the current version. A voluntary classification framework for publishers to build policy on, not a mandate.
- DAIU
- Declare AI Usage specification, version 1.0, draft status — open and voluntary, maintained by the Responsible AI Usage Foundation, Inc.
- ai-label.org
- Three-label public-domain icon scheme (No AI Used / Assisted by AI / Made with AI). Voluntary; the project states it “must be tested through real-world use and may contain biases.”
- Earned Trust (AIast)
- Version 1.6.3, deposited on Zenodo. This is the author's own standard — see the note below.
Disclosure: one of these eight is mine
Earned Trust (AIast) is my own standard. I compiled this comparison, and I am the author of one of the things being compared. It is included because it is a documented, publicly archived proposal in this space, and leaving it out would have been its own kind of dishonesty — but it does not belong in the same category as the others and I am not presenting it as though it does. It has been adopted by no journal and endorsed by no body. STM, COPE, ICMJE, Elsevier and Springer Nature govern real submissions; DAIU and ai-label.org are voluntary schemes with organisations behind them; Earned Trust is one independent researcher's proposal, archived so it can be checked. Read its row as a proposal placed alongside the field, not as a peer of it.
This dashboard summarizes publicly available guidance as a research aid for authors and is not legal advice or an official statement from any listed organization. Policies change frequently — always confirm current requirements directly with the target journal.