{"id":"01f38556-f844-45a1-a872-e82fb75e5bae","arxiv_id":"2509.05390","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A human who directs, critically reviews, approves, and takes responsibility for an LLM-generated paper can legitimately count as its author, by the same criteria that make non-writing senior researchers authors.","lead":"This paper argues that a researcher can be a legitimate author of a paper even if she writes no words directly, as long as she guides the work, critically reviews the drafts, approves the final version, and takes responsibility for it, and that this applies just as well when the writing was done by an AI chatbot. A smart generalist should read it because it forces a clear choice: accept AI-assisted writing as a real form of authorship, or admit that the credit given to many n","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Premise 2's identity claim is contradicted by Section IV.3: Jones must vet more than Smith, so the relevance filter is asserted, not derived, and the analogy is graded without a threshold.","rationale":"The reader's weakest assumption identifies exactly the load-bearing issue: the relevance filter in Section III.2 is asserted, and the Section IV.3 concession about Jones's higher vetting burden constitutes a direct difference in the contributions themselves. My concern is the same one; the reader already weighed it into a CONDITIONAL verdict. I do not think the paper should be rejected outright because the argument is clearly presented, the disjunctive conclusion is defensible, and the authoriality of LLM users is a live normative question. However, the paper does not provide a principle for determining relevant respects or a threshold for graded similarity, so the central modus ponens is under-defended. The concrete test—an itemized comparison of Smith's and Jones's actions plus a search for a threshold—would settle whether premise 2 can be repaired. Since the reader's CONDITIONAL verdict already captures this uncertainty, no change to the verdict is warranted.","tokens_in":1022,"tokens_out":998,"duration_ms":76727,"concrete_test":"Construct an itemized comparison of Smith's and Jones's actions under the paper's stipulated conditions. For Smith: provide thesis/structure/citations to Charlie, review Charlie's draft, suggest improvements, verify argument quality and citations, approve. For Jones: provide same thesis/structure/citations to LLM, review LLM draft, suggest improvements, verify argument quality and citations, approve. Then add the real-world constraint from Section IV.3: due to LLM hallucination rates (e.g., 47% fabricated references), responsible verification of a draft generated by an LLM requires per-citation fact-checking or independent corroboration, whereas a human postdoc can be reasonably trusted to have checked references. If any action in Jones's list differs (e.g., Jones performs additional verification steps that Smith does not), the identity premise is false. Then assess whether the paper's '","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central inference (Section III.2) requires that Smith and Jones contribute 'in identical relevant respects,' so that the nature of the drafting entity (human postdoc vs. non-agent LLM) is irrelevant to authorship. This relevance filter is stipulated, not derived: the paper offers no principled criterion for which differences between the cases matter for authorship status. The paper itself undermines the identity claim in Section IV.3: because LLMs hallucinate at non-negligible rates (Bhattacharyya et al. 2023), Jones must engage with the manuscript more critically than Smith must with a postdoc's draft to responsibly take accountability. This is not a difference in social goods (IV.2) but a difference in the authorial contribution itself—specifically in the 'revising it critically' clause of the ICMJE criteria. The reply that Jones is thereby 'more authorial' (IV.3) abandons the exact analogy and supplies no threshold or principle for when a graded increase in critical review suffices for authorship. Without a threshold, premise 2 remains unjustified: the modus ponens does not go through, and the paper's stronger claim that Jones is an author is unsupported. The weaker disjunctive conclusion ('either Jones is an author or current criteria need revision') also depends on the same contested premise, so it is not independently secured.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper argues that a human researcher who uses an LLM to generate a complete research manuscript can nonetheless be an author, by analogy with a senior author who does not write any words. It introduces two parallel cases (Smith supervising a postdoc; Jones supervising an LLM), argues that Smith meets the ICMJE authorship criteria, claims that Smith and Jones contribute to their respective papers in the same relevant respects, and concludes by modus ponens that Jones is an author. The paper then replies to objections: denying premise 1, the disanalogy of social goods, the responsibility/hallucination problem, and the lack of moral agency in LLMs. It ends with a disjunctive conclusion: either LLM-assisted authorship is legitimate or current authorship norms require fundamental revision.","tokens_in":13224,"tokens_out":5167,"duration_ms":57501,"significance":"If the argument were successful, it would have direct implications for publication ethics, credit allocation, and the concept of authorship in biomedical and scientific research. The paper is clearly structured, takes major objections seriously, and is unusually candid about the limits of its own argument, including the empirical literature on LLM hallucination and the context-dependence of authorship norms. It also contains a transparent AI-use declaration. The central philosophical claim is timely and important, and the case-study method makes the argument accessible. However, the inference rests on at least two load-bearing assumptions that are asserted rather than adequately defended: the sufficiency of the ICMJE criteria and the identity of Smith's and Jones's contributions in the relevant respects. These need repair before the argument can be considered sound.","major_comments":[{"comment":"The justification of Premise 2 is undermined by the paper's own concession that Jones must engage more critically with LLM output than Smith must with a postdoc's draft. Section II stipulates that Smith and Jones provide the same parameters, receive the same draft, and provide the same feedback, but Section IV.3 says that 'in real cases, this may not strictly be true' and that Jones likely faces a higher burden due to LLM hallucination. If Jones must do more vetting to responsibly take responsibility, then her contribution is not identical in the relevant respect, and the exact analogy collapses. The reply that this makes Jones 'more authorial' (IV.3) shifts from a binary identity claim to a graded comparison without supplying a threshold. Consequently, the disjunctive conclusion ('either Jones is an author or current criteria require revision') is not independently secured: if Premise 2","section":"Section III.2 and Section IV.3"},{"comment":"The argument treats the four ICMJE criteria as sufficient for authorship, but this sufficiency is asserted rather than defended. The footnote offers two pragmatic reasons—mitigating unfair denial of credit and preventing strategic omission—but these do not establish conceptual sufficiency. If the criteria are only necessary, then Smith's satisfying them does not entail that Smith is an author, and Premise 1 fails. The paper does not engage with alternative authorship standards, such as a requirement of direct textual contribution or intentional authorship, except to dismiss them as contextually inapplicable. Since the entire modus ponens rests on this step, the sufficiency claim needs a more robust defense.","section":"Footnote 4 and Section III.1"},{"comment":"The treatment of the responsibility objection is ambiguous about whether the accountability criterion is necessary for authorship. The paper cites Levy (2025) to contest the ICMJE accountability requirement, but then continues to rely on Jones's responsibility-taking as part of what makes her 'more authorial.' If accountability is not required, the extra vetting burden is irrelevant to authorship; if it is required, then the case description must explicitly specify that Jones in fact performs the additional vetting beyond what Smith does. As written, the paper leaves the status of Premise 2 unclear and does not provide a determinate answer.","section":"Section IV.3"}],"minor_comments":[{"comment":"The sentence 'though we consider possible disanalogies in Section III.2' appears to be a cross-reference error; the disanalogies are considered in Section IV.2 (and IV.3).","section":"Section III.2"},{"comment":"The word 'artesanal' should be 'artisanal.'","section":"Section IV.4"},{"comment":"The paper uses 'senior author' to characterize Jones, but Jones is a sole author in the LLM case. The terminological relationship between 'senior authorship' and 'sole authorship' should be clarified early, since the analogy to Smith's co-authorship is not exact on its face.","section":"General"},{"comment":"The discussion of externalist reasons to consider LLMs as full-fledged authors is presented as 'somewhat weighty' but is then set aside as beyond scope. This digression could be trimmed or more clearly separated from the central argument, as it risks confusing the paper's main claim about the human user's authorship.","section":"Section IV.4"}],"recommendation":"major_revision","confidential_remarks":"The paper is within the scope of a bioethics or research ethics journal and addresses a timely question. The self-citations are frequent but are not used as premises for the central claim, so I do not see circularity. The main concerns are the undefended sufficiency of the ICMJE criteria and the collapse of the exact analogy in Section IV.3; these are fixable by reworking the argument into a direct sufficiency argument or by explicitly arguing that Jones's higher vetting burden satisfies the same threshold. I would be willing to review a revised version."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe thing to know about this paper: it's a clearly argued conceptual piece that makes a genuine contribution by drawing out the senior-author analogy. The authors separate three questions (can humans be authors, can LLMs, should they), and focus on the first. They build a transparent modus ponens: Smith the senior author is an author without writing a word; Jones the LLM user contributes the same way; therefore Jones is an author. The 'minimal senior authorship' framing is new and useful, and the objection-and-reply section is thorough—especially the treatment of mentorship and social goods, which they correctly keep separate from authorial status.\n\nThe soft spots are real but not disqualifying. The load-bearing premise is the relevance filter: that the nature of the drafting assistant (human postdoc vs. non-agent LLM) is irrelevant to the human's authorship. The paper asserts this rather than argues for it. The stress-test is right that Section IV.3 concedes Jones must vet more critically than Smith to responsibly take accountability, which is a difference in the 'revising critically' clause. The reply that Jones is thereby 'more authorial' shifts from exact analogy to graded similarity without a stated threshold. That said, the idealized cases are stipulated to be identical, and the concession only says real-world Jones has more work to do; it doesn't obviously destroy the analogy, but it does reveal that the relevance filter is doing all the work. Also, footnote 4 asserts that ICMJE criteria are sufficient for authorship, which is a contested normative assumption; the paper flags it but doesn't defend it beyond pragmatic reasons. A missing engagement with the updated ICMJE guidance on AI tools (which postdates the 2019 version they cite) is a minor gap.\n\nThe disjunctive conclusion ('either Jones is an author or authorship norms need revision') is weaker and still depends on the same relevance filter, so it doesn't independently rescue the modus ponens. But the paper's value isn't only in the proof; it's in the reframing, and the authors are honest about the normative assumptions.\n\nWho's it for: anyone working on research ethics, authorship policy, or AI ethics. It deserves serious peer review—not a desk reject. A good referee would push on the relevance filter and ask for a principled criterion. I'd bring it to a reading group.","headline":"Useful reframing of the LLM-authorship debate, but the key move—that the drafting entity's nature is irrelevant—is stipulated, not defended, and the paper's own vetting concession blurs the analogy.","tokens_in":13765,"tokens_out":2757,"would_cite":true,"duration_ms":27761,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A researcher can be the author of an LLM-written paper without writing a word, provided she guides the work, reviews it critically, and takes responsibility for it.","keywords":["research ethics","authorship criteria","large language models","senior authorship","LLM-assisted writing","ICMJE","publication ethics","philosophical argument"],"falsifier":"A systematic survey of journal authorship policies in bioethics and biomedical research could settle premise 1: if a substantial share of journals require authors to have directly drafted at least some text, then Smith fails the very criteria the analogy rests on, and Jones falls with her.","tokens_in":12823,"feed_emoji":"🤖","tokens_out":8229,"duration_ms":75594,"temperature":0.7,"pith_summary":"Many published papers list senior researchers as authors even though they never wrote any of the text; what matters is guiding the work, reviewing it critically, approving the final version, and taking responsibility. This paper argues that a researcher who does the same things with a large language model—prompting it, critiquing its draft, and approving the final text—is an author in exactly the same sense. The argument runs by analogy: if the senior author Smith is an author, and Jones's contribution to her LLM-assisted paper is identical in every respect that feeds into the usual authorship criteria, then Jones is an author too. The upshot is that LLM use, even to generate a complete draft, can be legitimate authorship under current norms; the only consistent alternative is to change those norms, which would strip authorship from many existing senior authors.","feed_headline":"LLM users can count as authors without writing a word","feed_subtitle":"Guiding, vetting, and vouching for an AI draft mirrors what senior authors already do, the paper argues.","key_machinery":"The argument turns on paired case studies—'Senior Author' Smith and 'LLM user' Jones—constructed so that the two researchers give identical instructions, receive the same draft, and give the same critical feedback. The load-bearing identity is the ICMJE four-criterion test, used as a filter that decides which differences between cases are 'relevant' to authorship: conception, critical revision, final approval, and accountability. Because the only difference between the cases is the nature of the assistant (human postdoc vs. LLM), and because that difference is not among the authorship-relevant categories, the two cases must receive the same verdict.","core_discovery":"The paper's central claim is that the prevailing, widely accepted authorship criteria (exemplified by the ICMJE four-part test) make authorship a function of conception, critical revision, final approval, and accountability—not of producing words. Since a paradigmatic senior author meets those criteria without writing, and since a researcher who guides an LLM can meet the same criteria in the same way, consistency requires recognizing the LLM user as an author. The paper formalizes this as a modus ponens: Smith is an author; if Smith is an author then Jones is an author; therefore Jones is an author. It also draws the contrapositive: if Jones is not an author, current authorship norms need f","pith_inferences":["If the analogy is right, the same logic should extend to other non-human drafting tools—automated analysis scripts, grammar engines, or future AI—so authorship attribution would track guidance and oversight rather than the mechanism of text production.","The argument predicts a testable pattern in empirical authorship-attribution studies: perceived authoriality should track conception, critical revision, and accountability more than word-level composition, so readers should rate Jones similarly to Smith when both roles are disclosed.","The paper leaves open a practical design question: journals may want a CRediT-style role that names 'conceptual guidance and critical revision of AI-generated text,' so that the human's authorial contribution is visible without requiring the LLM to be an author.","A deeper tension: if LLM output demands more vetting than a postdoc's draft, then Jones may need to do more authorial work than Smith to reach the same confidence; the paper treats this as making Jones 'more authorial,' but a stricter reading would make the two cases only similar in degree, not identical."],"forward_implications":["A researcher can be listed as (sole) author of a paper they never wrote a word of, as long as they conceived the project, critically reviewed LLM drafts, approved the final version, and accepted accountability.","If that conclusion is rejected, the rejection cannot stop with LLM users: current authorship criteria in medicine, bioethics, and many sciences would also stop covering senior authors who delegate drafting to a junior collaborator.","Irresponsible LLM use—submitting output without critical review and without accepting accountability—fails the authorship criteria, just as a senior author who rubber-stamps a postdoc's draft would fail them.","The paper's conclusion does not settle whether using LLMs is wise, fair, or ethical; those questions turn on separate empirical and normative considerations.","Formal role disclosure (such as CRediT statements) and AI-use statements are compatible with the argument and can keep credit honest while the human remains the author."],"supporting_citations":[{"why":"Supplies the four authorship criteria that make Smith an author and define which contributions count for the analogy.","marker":"ICMJE (2019 version)"},{"why":"Exhibits radical division of labor among thousands of co-authors, showing non-writing authorship is established practice.","marker":"COVIDSurg Collaborative and GlobalSurg Collaborative 2021"},{"why":"Provides the role-disclosure taxonomy Smith and Jones use to report their contributions transparently.","marker":"CRediT - Contributor Role Taxonomy n.d."},{"why":"Supplies evidence of LLM hallucination rates that grounds the responsibility objection to LLM-assisted authorship.","marker":"Bhattacharyya and colleagues (2023)"},{"why":"Argues that accountability is not required for authorship, undercutting a disanalogy between Smith's and Jones's responsibility burdens.","marker":"Levy (2025)"},{"why":"Supports the claim that in radically collaborative research no single author is wholly accountable for everything, normalizing Jones's situation.","marker":"Winsberg et al. 2014"},{"why":"Provides the intentional-communication theory behind the objection that LLM output is unauthored, which the paper rebuts.","marker":"Grice (1989)"},{"why":"States that researchers have a responsibility to vet LLM outputs, which is what lets a diligent Jones qualify as an author.","marker":"Porsdam Mann et al. 2023a"}],"fun_headline_variants":["LLM users may be authors, no writing needed","Senior author analogy justifies LLM authorship","If senior authors don't write, LLM users can too","Authorship isn't about writing, paper argues","Guide an AI draft? You might be an author"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The argument holds only if Smith and Jones are identical in every way that matters for authorship, and the paper defines 'what matters' by the ICMJE categories—so the difference between a human assistant and a non-agent LLM is treated as irrelevant without independent proof.","fun_headline_variants_meta":{"raw":{"variants":["LLM users may be authors, no writing needed","Senior author analogy justifies LLM authorship","If senior authors don't write, LLM users can too","Authorship isn't about writing, paper argues","Guide an AI draft? You might be an author"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000267,"raw_usage":{"total_tokens":1436,"prompt_tokens":717,"completion_tokens":719,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":461,"completion_tokens_details":{"reasoning_tokens":655}},"tokens_in":461,"tokens_out":719,"duration_ms":7013,"temperature":1.0,"reasoning_tokens":655,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T05:44:17.043060+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A systematic survey of journal authorship policies in bioethics and biomedical research could settle premise 1: if a substantial share of journals require authors to have directly drafted at least some text, then Smith fails the very criteria the analogy rests on, and Jones falls with her.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the four authorship criteria that make Smith an author and define which contributions count for the analogy."}],"review_version":1}