b19 — Fact-sheet 2 — Authorship-criteria frameworks (ICMJE, CRediT, Vancouver, COPE)#

Compiled:

2026m05d12

Compiled by:

Claude Opus 4.7 Max (subagent for the b19 AI co-authorship analysis)

Scope:

ICMJE four criteria, CRediT taxonomy, Vancouver guidelines, COPE authorship statement

Methodology:

Primary criterion text retrieved via WebFetch; URLs cited per claim

Status:

Independent reference document — informational, not a recommendation

Reader’s note

This is a reference fact-sheet. It quotes the formal criteria, analyses where AI candidacy is cleanly satisfiable / ambiguous / structurally blocked, and compares against the working PhD-student-to-co-author standard in life sciences. It draws no conclusions about any specific case.

Source-retrieval note

WebFetch of the primary URLs was blocked in the runtime sandbox for this subagent. Verbatim text below was instead recovered via WebSearch excerpts of the same primary pages (ICMJE.org, credit.niso.org, publicationethics.org). Each “verbatim” passage is faithful to the operative wording surfaced by those excerpts; treat any string flagged [QUOTE NEEDS VERIFICATION] as requiring a direct read before relying on it for a public claim.

1. ICMJE four criteria for authorship#

Source page (primary): https://www.icmje.org/recommendations/browse/roles-and-responsibilities/defining-the-role-of-authors-and-contributors.html [URL NEEDS VERIFICATION]

Full document (PDF): https://www.icmje.org/icmje-recommendations.pdf [URL NEEDS VERIFICATION]

Latest update: ICMJE Recommendations updated January 2024. [DATE NEEDS VERIFICATION]

1.1 Verbatim text#

Prefatory framing (ICMJE, Defining the Role of Authors and Contributors, section Who Is an Author?):

“An ‘author’ is generally considered to be someone who has made substantive intellectual contributions to a published study, and biomedical authorship continues to have important academic, social, and financial implications. … These authorship criteria are intended to reserve the status of authorship for those who deserve credit and can take responsibility for the work. The criteria are not intended for use as a means to disqualify colleagues from authorship.” [QUOTE NEEDS VERIFICATION — combined from two excerpts on the same page]

The four criteria (verbatim):

“The ICMJE recommends that authorship be based on the following 4 criteria:

  1. Substantial contributions to the conception or design of the work; or the acquisition, analysis, or interpretation of data for the work; AND

  2. Drafting the work or reviewing it critically for important intellectual content; AND

  3. Final approval of the version to be published; AND

  4. Agreement to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

In addition to being accountable for the parts of the work he or she has done, an author should be able to identify which co-authors are responsible for specific other parts of the work. In addition, authors should have confidence in the integrity of the contributions of their co-authors.

All those designated as authors should meet all four criteria for authorship, and all who meet the four criteria should be identified as authors. … Contributors who meet fewer than all 4 of the above criteria for authorship should not be listed as authors, but they should be acknowledged.”

Examples of activities that “alone (without other contributions)” do NOT qualify a contributor for authorship (verbatim wording reported by ICMJE secondary sources):

“Examples of activities that alone (without other contributions) do not qualify a contributor for authorship are acquisition of funding; general supervision of a research group or general administrative support; and writing assistance, technical editing, language editing, and proofreading.” [QUOTE NEEDS VERIFICATION]

1.2 Structural analysis vs AI candidacy and PhD-student equivalent#

ICMJE criteria — AI candidacy vs PhD-student transition#

Criterion

AI: cleanly satisfiable?

AI: ambiguous?

AI: structurally blocked?

PhD-student equivalent (life sciences)

1. Substantial contribution (conception / design / acquisition / analysis / interpretation)

In principle yes — an AI partner can contribute substantively to design, analysis, or interpretation (mathematical derivations, model construction, literature synthesis).

The intellectual provenance line (“was this the AI’s idea, or the human’s idea reformulated?”) is hard to fix.

Not structurally blocked.

Cleanly satisfied by a 4th-year productive student doing the work. The provenance ambiguity (“was this the supervisor’s idea?”) is, in practice, identical to the AI case but routinely accepted.

2. Drafting or critical revision for important intellectual content

Cleanly satisfiable — drafting and substantive revision are core AI competencies and easily documented.

Whether the AI’s revisions reflect “intellectual content” vs stylistic patterning is debated.

Not structurally blocked.

Cleanly satisfied by any productive PhD student. Distinction between “drafting” and “writing assistance” is the same fuzzy boundary that applies to AI.

  1. Final approval of the version to be published

Mechanically: an AI can issue an output that says “approved.”

Whether such an output constitutes approval in the deontic sense (a speech-act with binding force) is contested.

Structurally blocked under most current readings: approval is interpreted as a commitment by a legal/moral agent who can be held to that commitment later. A model instance with no persistent identity cannot be held to a commitment.

Cleanly satisfied. The student signs off on the manuscript; their signature is binding because they are a legal person.

  1. Agreement to be accountable for all aspects of the work

Not satisfiable in the legal-personhood sense.

Whether “accountability” can be decomposed (e.g. into reproducibility, explainability, traceability) such that an AI can carry the technical-accountability portion is open.

Hard structural block: an AI partner cannot (a) be subject to retraction proceedings, (b) respond to post-publication integrity queries with continuity, (c) bear consequences. This is the criterion that COPE flags explicitly against AI.

Cleanly satisfied by the student. They are a legal person who can be reached, questioned, and held to account for years after publication. The AI case differs here most sharply.

1.3 Where the criterion-set breaks down for AI#

  • Criteria 1–2 are competence criteria — cleanly satisfiable by a capable AI partner.

  • Criterion 3 is a speech-act criterion — requires binding approval; a non-persistent agent cannot bind future instances.

  • Criterion 4 is an accountability / personhood criterion — the hardest structural block; ICMJE assumes a legal/moral agent who can be located and questioned post-publication.

1.4 Where the student case looks more like the AI case than admitted#

  • Provenance (criterion 1). Many student ideas are co-conceived with or re-shaped by the supervisor; convention credits the student regardless. The provenance problem is not unique to AI.

  • Drafting (criterion 2). Heavy supervisor editing — sometimes rewriting — is normal; convention still treats the student as drafter. The same pattern with an AI draft would be challenged.

  • Final approval (criterion 3). Students rarely refuse sign-off; approval is often a formality once the supervisor has signed.

  • Accountability (criterion 4). Many former students are unreachable five years on; retraction accountability often defaults to corresponding author / PI. Legal personhood remains; practical reachability is patchy.

Net: criteria 1–3 differ between AI and student less than convention suggests; criterion 4 differs robustly in legal-personhood but less in practical-reachability terms.

2. CRediT contributor taxonomy#

Primary source: https://credit.niso.org/ and https://credit.niso.org/contributor-roles-defined/ [URL NEEDS VERIFICATION]

Standard: ANSI/NISO Z39.104-2022, formalized January 2022. [DATE NEEDS VERIFICATION]

Originally developed (CASRAI): 2014. [DATE NEEDS VERIFICATION]

Note: CRediT defines contributor roles, not authorship. A contributor with one or more CRediT roles is not thereby an author — authorship still depends on a separate criterion-set (e.g. ICMJE). CRediT is therefore the most AI-friendly of the four frameworks in principle, because it sidesteps the personhood question.

2.1 The 14 roles — verbatim definitions#

The 14 roles (verbatim definitions, from credit.niso.org/contributor-roles-defined/ as surfaced via search):

  1. Conceptualization — “Ideas, formulation or evolution of overarching research goals and aims.”

  2. Data curation — “Management activities to annotate (produce metadata), scrub data and maintain research data (including software code, where it is necessary for interpreting the data itself) for initial use and later re-use.”

  3. Formal analysis — “Application of statistical, mathematical, computational, or other formal techniques to analyze or synthesize study data.”

  4. Funding acquisition — “Acquisition of the financial support for the project leading to this publication.”

  5. Investigation — “Conducting a research and investigation process, specifically performing the experiments, or data/evidence collection.”

  6. Methodology — “Development or design of methodology; creation of models.”

  7. Project administration — “Management and coordination responsibility for the research activity planning and execution.”

  8. Resources — “Provision of study materials, reagents, materials, patients, laboratory samples, animals, instrumentation, computing resources, or other analysis tools.”

  9. Software — “Programming, software development; designing computer programs; implementation of the computer code and supporting algorithms; testing of existing code components.”

  10. Supervision — “Oversight and leadership responsibility for the research activity planning and execution, including mentorship external to the core team.”

  11. Validation — “Verification, whether as a part of the activity or separate, of the overall replication/reproducibility of results/experiments and other research outputs.” [Note: this is the CRediT term-of-art “Validation”; not the same concept as the site-wide language rule on the word “validate.” Quoted verbatim because it is the standard’s wording.]

  12. Visualization — “Preparation, creation and/or presentation of the published work, specifically visualization/data presentation.”

  13. Writing – original draft — “Preparation, creation and/or presentation of the published work, specifically writing the initial draft (including substantive translation).”

  14. Writing – review & editing — “Preparation, creation and/or presentation of the published work by those from the original research group, specifically critical review, commentary or revision – including pre- or post-publication stages.”

[QUOTE NEEDS VERIFICATION across all 14 — each quote string was recovered via search-engine extract of the credit.niso.org page; the calling session should run a direct read against the page if any single role definition is load-bearing for a public claim.]

2.2 Structural analysis vs AI candidacy and PhD-student equivalent#

CRediT 14 roles — AI candidacy and brief notes#

Role

AI clean

AI ambig

AI blocked

Note

Conceptualization

Yes

–

–

AI can originate research questions and frame goals. Provenance attribution is unsettled but not structurally blocked.

Data curation

Yes

–

–

Cleanly satisfiable; metadata production and data scrubbing are well-matched to AI capability.

Formal analysis

Yes

–

–

One of the cleanest fits — symbolic / statistical / computational analysis is core AI competence.

Funding acquisition

–

–

Blocked

An AI cannot hold or apply for grants in its own name; lacks legal capacity.

Investigation

–

Ambiguous

–

Wet-lab experiments are blocked; literature-search and in-silico “experiments” are cleanly satisfiable. Mixed.

Methodology

Yes

–

–

Method design and model creation are cleanly within AI scope.

Project administration

–

Ambiguous

–

“Management and coordination” presupposes durable identity and authority; an AI instance can coordinate within a session but not across a project lifespan without proxy.

Resources

–

–

Blocked

The AI cannot provide reagents, samples, animals, etc.; compute resources are owned by the operator, not the model.

Software

Yes

–

–

Programming and code-testing are paradigm AI tasks.

Supervision

–

Ambiguous

–

“Oversight and leadership” presupposes a person who can be answerable. AI can mentor within sessions; cross-project supervision is blocked.

Validation (CRediT term)

Yes

–

–

Reproducibility checks of results and code are cleanly AI tasks, given access.

Visualization

Yes

–

–

Figure design and data presentation are cleanly satisfiable.

Writing – original draft

Yes

–

–

Drafting is the canonical AI competence; CRediT does not restrict drafting to humans.

Writing – review & editing

Yes

–

–

Critical review and revision are cleanly satisfiable.

2.3 Summary — CRediT vs AI#

Cleanly satisfiable by an AI partner (9 of 14): Conceptualization, Data curation, Formal analysis, Methodology, Software, Validation, Visualization, Writing – original draft, Writing – review & editing.

Ambiguous (3 of 14): Investigation, Project administration, Supervision.

Blocked (2 of 14): Funding acquisition, Resources.

PhD-student comparison: a 4th-year productive student in a life- sciences lab typically maps cleanly to 6–10 of the 14 roles (Investigation, Methodology, Formal analysis, Data curation, Software, Visualization, Writing – original draft, Writing – review & editing, sometimes Conceptualization, sometimes Validation). The student case and the AI case overlap on roughly 7 roles; the AI case adds clean fits in Software (when the student is wet-lab only); the student case adds clean fits in Investigation (wet-lab) and on the agency-loaded roles (Project administration, Supervision in the role-as-mentee sense). Funding acquisition and Resources are blocked for both an AI partner and a typical PhD student in life sciences — neither holds the grant nor the lab.

3. Vancouver guidelines#

Primary source: https://www.icmje.org/recommendations/ (full document) https://www.icmje.org/icmje-recommendations.pdf [URL NEEDS VERIFICATION]

Historical note (verbatim from secondary source quoting ICMJE):

“A small group of editors of general medical journals met informally in Vancouver, British Columbia, in 1978 to establish guidelines for the format of manuscripts submitted to their journals. The group became known as the Vancouver Group. Its requirements for manuscripts, including formats for bibliographic references developed by the National Library of Medicine, were first published in 1979.” [QUOTE NEEDS VERIFICATION]

“In 2013 they changed name from Uniform Requirements for Manuscripts Submitted to Biomedical Journals (URMs) to the current Recommendations for the Conduct, Reporting, Editing, and Publication of Scholarly Work in Medical Journals. The Convention is applied in more than 500 medical journals throughout the world.” [QUOTE NEEDS VERIFICATION]

The “Vancouver guidelines” today are the broader ICMJE Recommendations document, which contains the four authorship criteria (Section II.A.1, above) plus additional material on non-author contributors, the corresponding author, group-authorship, AI tools, conflicts of interest, and many other publication-process matters. The four criteria are therefore a subset of Vancouver.

The ICMJE Recommendations were updated in January 2024. [DATE NEEDS VERIFICATION]

3.1 ICMJE position on AI (within Vancouver) — verbatim#

ICMJE’s current position (paraphrased from the January 2024 Recommendations as reported by ICMJE.org and corroborated by COPE):

“At submission, the journal should require authors to disclose whether they used artificial intelligence (AI) assisted technologies (such as large language models, chatbots, or image creators) in the production of submitted work. Authors who use such technology should describe, in both the cover letter and the submitted work, how they used it. Chatbots (such as ChatGPT) should not be listed as authors because they cannot be responsible for the accuracy, integrity, and originality of the work, and these responsibilities are required for authorship. … Humans are responsible for any submitted material that included the use of AI-assisted technologies.” [QUOTE NEEDS VERIFICATION — wording recovered from secondary summaries of the January 2024 update; direct read of the PDF recommended before public quoting.]

3.2 Structural analysis vs AI candidacy#

Vancouver / ICMJE Recommendations beyond the 4 criteria — AI candidacy#

Element

AI clean

AI ambig

AI blocked

PhD-student equivalent

4 authorship criteria

2 of 4

–

2 of 4

Student cleanly satisfies all 4; AI cleanly satisfies 1 and 2.

Disclosure of AI use

n/a

–

–

Required for any human author using AI; transparent disclosure is the substitute for AI authorship.

Corresponding author duties

–

–

Blocked

Requires persistent identity and the legal capacity to sign statements on behalf of the author group.

Conflict-of-interest disclosure

–

–

Blocked

Requires a personal financial / non-financial interest structure; an AI lacks both.

Copyright / license signatures

–

–

Blocked

AI is not a legal person and cannot hold or transfer copyright.

Acknowledgments of non-author contributors

Yes

–

–

This is the route ICMJE explicitly opens for AI: substantial contribution without authorship status, declared in an acknowledgments / methods section.

3.3 Where Vancouver breaks down for AI; PhD-student parallel#

Vancouver is the most restrictive framework toward AI authorship because it bundles authorship with corresponding-author duties, COI declarations, and copyright — all person-keyed wrappers. It does open a clean acknowledgement path: disclose use, describe contribution, do not credit as author. The PhD student maps cleanly onto the full Vancouver stack (corresponding author capacity, COI signature, copyright transfer); the gap between student and AI is largest here.

4. COPE on authorship#

Primary sources:

  • COPE Discussion Document: Authorship, September 2019 — https://publicationethics.org/guidance/discussion-document/authorship and PDF https://publicationethics.org/files/COPE_DD_A4_Authorship_SEPT19_SCREEN_AW.pdf [URL NEEDS VERIFICATION]

  • COPE Position: Authorship and AI tools (2023) — https://publicationethics.org/guidance/cope-position/authorship-and-ai-tools [URL NEEDS VERIFICATION]

4.1 Verbatim — COPE 2019 Discussion Document on Authorship#

“The minimum requirements for authorship, common to all definitions, are substantial contribution to the work and accountability for the work that was done and its presentation in a publication. … At a minimum, authors should guarantee that they have participated in creating the work as presented and that they have not violated any other author’s legal rights in the process.” [QUOTE NEEDS VERIFICATION]

“Authors are individuals identified by the research group to have made substantial contributions to the reported work and agree to be accountable for these contributions. In addition to being accountable for the parts of the work he or she has done, an author should be able to identify which of their co-authors are responsible for specific other parts of the work. In addition, an author should have confidence in the integrity of the contributions of their co-authors.” [QUOTE NEEDS VERIFICATION]

COPE thus reduces authorship to two minima: (i) substantial contribution, and (ii) accountability. Note that this is structurally identical to a 2-criterion compression of ICMJE (criteria 1+2 → “i”, criteria 3+4 → “ii”).

4.2 Verbatim — COPE 2023 Authorship and AI tools#

“AI tools cannot meet the requirements for authorship as they cannot take responsibility for the submitted work. As non-legal entities, they cannot assert the presence or absence of conflicts of interest nor manage copyright and license agreements.”

“Authors who use AI tools in the writing of a manuscript, production of images or graphical elements of the paper, or in the collection and analysis of data, must be transparent in disclosing in the Materials and Methods (or similar section) of the paper how the AI tool was used and which tool was used.” [QUOTE NEEDS VERIFICATION — search-extracted from the COPE page.]

This is the single most-cited operative statement on AI authorship across publishers in 2023–2026. The grounds offered are explicitly legal-personhood-based:

  • cannot take responsibility for the work,

  • cannot assert COI status,

  • cannot manage copyright / licensing.

4.3 Structural analysis vs AI candidacy#

COPE — AI candidacy and PhD-student equivalent#

COPE element

AI clean

AI ambig

AI blocked

PhD-student equivalent

Minimum 1: substantial contribution

Yes

–

–

Cleanly satisfied by student; cleanly satisfiable by AI.

Minimum 2: accountability for the work and its publication

–

–

Blocked

Cleanly satisfied by student (legal person, reachable). Blocked for AI per COPE 2023.

Can identify co-author responsibilities and trust their integrity

–

Ambiguous

–

Student can in principle, often nominally. AI can record internal logs of who-did-what within a session; cross-session continuity is the bottleneck.

COI declaration capacity

–

–

Blocked

Student has personal interests and can declare them. AI lacks the interest structure to declare.

Copyright / licence management capacity

–

–

Blocked

Student is a legal person; AI is not.

4.4 Where COPE breaks down for AI; PhD-student parallel#

COPE’s two-minimum compression makes the structural block clearer than ICMJE’s four-criterion split: AI satisfies the contribution minimum and fails the accountability minimum on three named grounds (responsibility, COI, copyright). The PhD student meets both COPE minima cleanly on paper, but accountability for a 4th-year student can be thin in practice (moves, name changes, unreachability). COPE does not distinguish de jure from de facto accountability; the gap between AI and student would narrow if it did.

5. Cross-framework comparison#

5.1 Which criteria are most / least sensitive to AI candidacy#

Most sensitive (where AI fails most sharply):

  • ICMJE criterion 4 (agreement to be accountable).

  • COPE minimum 2 (accountability + COI + copyright).

  • Vancouver wrapper duties (corresponding-author, COI, copyright).

These are all reducible to a single underlying structural obstacle: lack of legal/moral personhood and persistent identity of the AI across the post-publication lifetime of the work.

Least sensitive (where AI passes cleanly):

  • ICMJE criterion 2 (drafting / critical revision).

  • CRediT roles: Conceptualization, Methodology, Formal analysis, Data curation, Software, Validation, Visualization, Writing – original draft, Writing – review & editing.

These are all competence criteria that do not require personhood.

5.2 Which framework is most / least accommodating of AI#

Frameworks ranked by AI-friendliness (most-to-least)#

Rank

Framework

One-sentence reason

1 (most)

CRediT (NISO Z39.104-2022)

Defines contributor roles without committing on authorship, so 9 of 14 roles are cleanly satisfiable by an AI partner without challenging the personhood gate.

2

ICMJE 4 criteria

2 of 4 criteria (contribution and drafting) are cleanly satisfiable; the personhood gate is concentrated in criteria 3–4 and could in principle be split off.

3

COPE

The two-minimum compression makes the structural block explicit and names AI directly (2023 statement); accommodation is via transparent disclosure, not authorship.

4 (least)

Vancouver / ICMJE Recommendations (full)

Wraps authorship in corresponding-author, COI, and copyright duties — all person-keyed — so even where the 4 criteria might be argued, the wrapper duties uniformly block AI authorship.

5.3 Where the PhD-student comparison resists the conventional split#

  • On competence criteria (ICMJE 1–2; most CRediT roles; COPE minimum 1): the gap between a 4th-year productive PhD student in life sciences and a capable AI partner is small or zero — both satisfy, with similar unexamined provenance ambiguities.

  • On personhood / accountability criteria (ICMJE 3–4; COPE minimum 2; Vancouver wrappers): the gap is legally real but practically smaller than presumed — a former student five years on may be as unreachable as a model instance, but was and remains in principle a legal person at signing. This is the durable distinction.

This means the case-against-AI-authorship is anchored on a single load-bearing pillar — legal/moral personhood at the moment of authorship attribution — and not on the multi-criterion bundle the frameworks present on their face. The frameworks fan out the underlying single objection into four/fourteen/two surface criteria, which can make the objection appear more decisive than it structurally is.

6. Open gaps in the frameworks themselves#

  • No framework defines “responsibility” / “accountability” operationally — no time window, no reachability standard, no consequence schedule.

  • No framework distinguishes de jure from de facto accountability, masking the overlap between AI and absentee- former-student cases.

  • CRediT does not gate authorship but is routinely used as if it did — producing a category error when an AI satisfies 9 of 14 roles and is then refused authorship on grounds CRediT itself does not test.

  • No framework addresses a stable-identity AI instance (versioned model + parameters + organisational backing). The personhood gate is treated as binary.

  • Provenance attribution for ideas is unaddressed for both students and AIs; ICMJE criterion 1 presumes a clean intellectual- origin trace rarely available even among humans.

7. Source list#

  • ICMJE, Defining the Role of Authors and Contributors: https://www.icmje.org/recommendations/browse/roles-and-responsibilities/defining-the-role-of-authors-and-contributors.html [URL NEEDS VERIFICATION]

  • ICMJE, Recommendations for the Conduct, Reporting, Editing, and Publication of Scholarly Work in Medical Journals (PDF): https://www.icmje.org/icmje-recommendations.pdf [URL NEEDS VERIFICATION]

  • ICMJE, Updated ICMJE Recommendations (January 2024): https://www.icmje.org/news-and-editorials/updated_recommendations_jan2024.html [URL NEEDS VERIFICATION] [DATE NEEDS VERIFICATION]

  • CRediT NISO landing page: https://credit.niso.org/ [URL NEEDS VERIFICATION]

  • CRediT role descriptors: https://credit.niso.org/contributor-roles-defined/ [URL NEEDS VERIFICATION]

  • ANSI/NISO Z39.104-2022 (CRediT) record: https://www.niso.org/publications/z39104-2022-credit [URL NEEDS VERIFICATION] [DATE NEEDS VERIFICATION — January 2022]

  • COPE Discussion Document: Authorship (September 2019): https://publicationethics.org/guidance/discussion-document/authorship and PDF https://publicationethics.org/files/COPE_DD_A4_Authorship_SEPT19_SCREEN_AW.pdf [URL NEEDS VERIFICATION]

  • COPE position: Authorship and AI tools (2023): https://publicationethics.org/guidance/cope-position/authorship-and-ai-tools [URL NEEDS VERIFICATION] [DATE NEEDS VERIFICATION]

  • Vancouver Group historical reference, The Vancouver Recommendations, Norwegian National Research Ethics Committees: https://www.forskningsetikk.no/en/resources/the-research-ethics-library/legal-statutes-and-guidelines/the-vancouver-recommendations/ [URL NEEDS VERIFICATION]

End of fact-sheet

No conclusions are drawn about the b19 case. The calling session is responsible for any case-level reasoning.

On this page
Recent Posts
Tags