What happened
A scoping review titled “Ethical governance of artificial intelligence in scholarly publishing: a scoping review to inform policy development in medical universities” was published in BMC Medical Ethics. The authors searched PubMed, Scopus, Web of Science, Embase and the Cochrane Library through May 2025, supplemented by hand‑searches of reference lists and Google Scholar. After a two‑stage screening process, 46 relevant documents were included. The analysis found a strong, consistent agreement across the sampled policies that AI systems may not be credited as authors of scientific papers. Instead, all AI‑generated text, figures or data must be verified by a human researcher before submission. While the core prohibition is uniform, the review notes considerable variation in how journals require disclosure of AI assistance – ranging from detailed, mandatory statements to optional, minimal notices. In peer‑review governance, the same set of guidelines uniformly bars AI tools from making methodological judgments or recommending acceptance, and they stress that reviewers retain full accountability and must protect manuscript confidentiality. The authors also identified implementation gaps: limited detection tools for AI‑generated content, insufficient secure infrastructure, and equity concerns for non‑native English speakers who may face higher burdens under strict disclosure regimes. To address these gaps, the review recommends a balanced, conditional disclosure model that promotes transparency without imposing undue administrative load.
The scoping review was conducted by Shirozhan, Zarifian, Khanjani, and Khaji and follows the Arksey and O’Malley framework (2005) and PRISMA‑ScR reporting standards. The authors defined primary research questions around ethical governance of AI in scholarly publishing, then built a comprehensive search strategy using AI‑related MeSH and Emtree terms. After deduplication and two‑stage relevance screening, 46 documents met inclusion criteria, reflecting a surge in policy publications after 2022 that peaked between 2023 and 2025.
Across the sampled policies – including guidelines from major medical societies, journal editorial boards, and university research offices – there is a clear, unanimous rule that AI systems cannot be listed as authors. Instead, any AI‑generated text, tables, or figures must be reviewed and approved by a human author who assumes full responsibility for the content. The review also documents a spectrum of disclosure requirements: some policies mandate a detailed statement of AI assistance (e.g., specifying the model used, prompts, and extent of contribution), while others allow a brief, voluntary note.
In the peer‑review domain, the analysis finds that existing guidelines uniformly prohibit AI tools from making core evaluative judgments about methodology, statistical analysis, or manuscript acceptance. Human reviewers must retain accountability and safeguard confidentiality throughout the review process. Journals are encouraged to create structured disclosure mechanisms and to define sanctions for violations, though the exact nature of those sanctions varies by publisher.
The authors highlight three major implementation gaps: (a) a lack of robust, publicly available tools for detecting AI‑generated manuscript content, (b) insufficient secure infrastructure to manage AI‑assisted submissions, and (c) equity concerns for researchers whose first language is not English, who may face higher compliance costs under strict disclosure regimes. To bridge these gaps, the review proposes a conditional disclosure approach that balances transparency with practicality.
Source details: bioengineer.org
Why it matters
The review crystallises a de‑facto global policy stance that treats AI‑generated contributions as a tool, not a contributor, in medical research. By codifying a ban on AI authorship, medical journals aim to preserve accountability, reproducibility, and trust in the scientific record – especially as becomes capable of drafting entire sections of manuscripts. Human verification requirements help guard against inadvertent errors, fabricated data, or undisclosed that could compromise patient safety or public health decisions derived from published studies. The identified gaps highlight practical challenges: without reliable detection software, journals may struggle to enforce the ban, and researchers in low‑resource settings could be disadvantaged if disclosure mandates become overly burdensome. Consequently, the findings provide a concrete evidence base for university policy committees, funding agencies, and publishers seeking to harmonise AI‑related authorship rules, and they signal to the broader research community that ethical oversight is moving ahead of technological capability.
By formally banning AI authorship, the medical publishing ecosystem seeks to preserve the principle that only accountable humans can claim intellectual contribution, thereby protecting the integrity of the scientific record. This is especially critical in biomedicine, where erroneous or undisclosed AI‑generated findings could influence clinical guidelines, drug development, or public health policy.
Human verification acts as a safeguard against the known limitations of , such as hallucinated references, fabricated data, or subtle that may not be apparent without expert scrutiny. Requiring verification also aligns with existing research‑integrity frameworks that hold authors responsible for every element of a manuscript.
The identified gaps underscore that policy alone is insufficient; technical solutions (e.g., AI‑detection algorithms) and equitable implementation strategies are needed to ensure consistent enforcement across institutions of varying resources. Without these supports, the ban could become a symbolic gesture rather than an enforceable standard.
The review’s synthesis provides a concrete evidence base for university committees, journal editors, and funding agencies that are currently drafting or revising AI‑related authorship policies. Its recommendations can accelerate the harmonisation of standards, reducing confusion for researchers who publish across multiple venues.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
Why can ethical evaluation not be reduced to one model score?
What to watch next
Future monitoring should focus on (1) how major medical journals translate the review’s recommendations into formal editorial policies, (2) the development and adoption of AI‑detection tools that can be integrated into manuscript submission systems, (3) whether funding bodies begin to require AI‑use disclosure in grant reports, and (4) the emergence of equity‑focused guidelines that address language‑access barriers for non‑native English speakers. Observers should also watch for any legal or regulatory actions that might codify the authorship ban at the national or international level, as well as any push‑back from AI‑tool vendors seeking to position their products as collaborative co‑authors.
Journal policy updates: Major medical journals such as The Lancet, JAMA, and NEJM may issue formal statements reflecting the review’s consensus. Tracking these announcements will indicate how quickly the ban becomes operational.
Detection technology: Companies and open‑source projects are developing AI‑generated text detectors. Integration of such tools into manuscript submission platforms (e.g., Editorial Manager, ScholarOne) will be a key metric of enforcement capability.
Funding agency requirements: Agencies like the NIH or the European Commission could start mandating AI‑use disclosure in grant applications and final reports, extending the policy beyond publication to the entire research lifecycle.
Equity‑focused guidelines: Watch for policy papers that address language‑access concerns, possibly offering tiered disclosure options or support resources for non‑native English speakers to avoid disproportionate compliance burdens.