Use of AI is now outpacing disclosure in manuscript preparation for major scholarly publications. This is a fact that publishers and journals must come to grips with and take steps to address to manage accountability from authors. As new technologies and tools continue to emerge, journals need to decide what they will and won’t be allowing authors to use during manuscript preparation. Many large publishers and societies now need to develop in-depth policies to deal with questions of authorship and generative AI.
Who Is the Author?
AI is no longer just a writing tool. Emerging agentic systems can plan, act, and execute workflows. They can generate manuscripts, create identities, submit papers, and respond to feedback. Paper mills are evolving into operational systems.
Publishers and editorial offices are being asked to evaluate both the research quality and the appropriate use of AI. The key question becomes: Is there a real, accountable human behind the submission? And then, when should you start going down the rabbit hole, and when should you escalate what seems to be a case of misconduct?
What Is the Right Way to Address Generative AI?
The use of generative AI is controversial among journals with some taking hardline approaches, while other publishers offer a more nuanced approach. For example, the Journal of the American Veterinary Medical Association does not allow the use of any generative AI programs in manuscript preparation. On the other hand, Taylor & Francis have a slightly laxer policy for the use of generative AI with requirements. Generative AI can be used by authors according to their policy, but an AI tool cannot be listed as an author. Additionally, authors must clearly acknowledge any use of generative AI within the article or book when submitting.
There are unequal levels of access to technology, and it is hard to effectively enforce any policy. Stakeholders want to know that humans were involved in a meaningful way in any publication submissions. Multiple studies show that researchers use AI for many tasks, but AI disclosure is low sometimes due to fear of bias along with inconsistent guidelines by many journals regarding the use of tools.
How Should You Formulate Your AI Policy?
There is no right or wrong way to craft an AI policy, but a representative at AIP Publishing outlined their strategy at the recent CSE conference in Durham, NC.
AIP Publishing launched its first AI ethics policy in 2026. They started by interviewing journal editors about AI use in their fields to help shape the disclosure requirements. It was determined that AI disclosure needed to be somewhat broad, with opportunities to be tailored to different individual fields and journals. Providing clear examples relevant to an author’s work, as outlined in the stated policy on the website, was also important. This helps reduce the stigma around AI use by authors by acknowledging the value of AI as an extension of human capabilities while also creating clear boundaries that authors cannot cross when using AI.
As a result of this research, AIP developed this policy for authors, editors, and reviewers. It provides clear rules with regard to using generative AI for images and states rules against AI tools being listed as authors. However, it offers opportunities for authors to use certain tools as long as they are fully disclosed by the authors in transparent ways. It also provides best practices for the authors to review in case of questions about their AI usage.
What to Do if You Suspect Potential Misconduct?
What is the best course of action if a journal suspects journal misconduct? Based on discussions with editors and publishers, the use of AI tools as authors appears to be considered the biggest potential case of AI misconduct and a red flag that the manuscript is part of a paper mill. If you suspect a paper has been generated by AI and includes fake author listings, when should you investigate further and when should you report it?
You should start by watching for noninstitutional emails, email domains that do not match stated affiliations, recently created domains, and multiple authors sharing similar noninstitutional emails. These are all usually signs of potential paper mill activity. This includes same or similar numerical endings for all authors in email domains. Also, look for multiple authors having the same email address or authors having linked emails with secondary emails that match another author on the submission. Of course, these are just warning signals and do not automatically denote fraud. Noting any of the above listed issues should not be cause for an obvious rejection by the editorial office, but it could be cause for concern.
Another surface signal is missing author metadata including missing affiliations, missing ORCID IDs, inconsistent author name formatting, affiliations that are difficult to verify online. If this is noted, the editorial office or a research integrity team can do a crosscheck. That crosscheck should include asking the following questions:
- Does the ORCID match the name, affiliation, and field?
- Is the record active and consistent?
- Do linked publications align with submission topics?
- Does the author’s institution exist and have a web presence?
- Is the author listed on institutional pages with the matching departments?
- Do the authors have previous publications in the field?
- Have they participated in conferences?
- Have they provided reviewer suggestions?
- Do the reviewer suggestions have similar email domain or institutional issues?
Conclusion
AI policies are becoming a necessity for scholarly journals and publishers. As more and more manuscripts are submitted using AI tools, it is important for journals to create clear boundaries for its usage in manuscript preparation. These clear boundaries can help clarify things for authors and eliminate the stigma surrounding the use of certain tools, while also creating a framework for investigating potential misconduct allegations. At the end of the day, these policies work as a trust exercise between publishers and authors. It is the author’s responsibility to be truthful with their disclosures, while it is the publisher’s responsibility to trust but verify.
