The 2025 COPE Retraction Guidelines: What Medical Authors Need to Know

Most medical authors spend very little time thinking about retraction policy until they receive an email that forces them to. That email might come because a reviewer spotted an inconsistency after publication, because a dataset was flagged by an integrity checking tool, because a co-author is under investigation at another institution, or because a journal has identified a cluster of papers from a submission agency and yours happened to land in the same group. When that moment arrives, the rules that matter most are the ones the Committee on Publication Ethics publishes, because most reputable journals cite COPE guidance as the basis for their own retraction procedures.

COPE released Version 3 of its retraction guidelines in August 2025, accompanied by a separate new document on expressions of concern. Both documents were updated to reflect patterns that the previous version, from 2019, could not have fully anticipated: the rise of paper mills operating at industrial scale, the spread of undisclosed AI use in manuscript preparation, and the procedural problem of handling dozens or hundreds of suspect papers from the same source simultaneously. These are not minor editorial adjustments. They formalize a set of practices that journals have been developing inconsistently for several years and bring them under a single, more detailed framework.

What COPE Is, and Why Its Guidelines Set the Standard

The Committee on Publication Ethics was founded in 1997 by a small group of medical journal editors who recognized that the field had no shared framework for handling misconduct cases. It started as a UK-based group and has grown into an international organization with over 13,000 member journals, spanning every academic discipline. Its membership includes major publishers (Elsevier, Springer Nature, Wiley, Taylor and Francis, BMJ, JAMA Network) as well as thousands of independent and society journals. Membership is voluntary but signals a commitment to following COPE's ethical standards, and many university libraries, research funders, and indexing databases now check COPE membership as part of their quality assessment.

COPE does not directly investigate cases or issue retractions itself. Its role is to publish guidelines, run training for editors, and provide case consultation. The guidelines are advisory documents that journals adopt and adapt into their own policies. In practice, when a medical journal retracts a paper, it typically cites COPE's retraction guidelines as the basis for the decision, and the editor will often have checked with COPE's flowchart resources before reaching a conclusion. Knowing what those guidelines say gives you a clearer picture of what is and is not discretionary when a journal contacts you about a potential problem.

The 2019 Version 2 of the retraction guidelines was already fairly detailed, but it predated several problems that have since become endemic: the industrialization of paper mills, the rapid adoption of generative AI tools by researchers, and the emergence of review manipulation at large enough scale to require coordinated responses across multiple journals at once. Version 3 addresses each of these directly.

What Triggers a Retraction Under the New Framework

The core principle of the retraction framework has not changed: retraction is a tool for correcting the scientific literature, not for punishing authors. Version 3 makes this explicit. But the practical list of what can trigger a retraction has expanded meaningfully.

The previous guidelines listed honest error, unreliable findings, redundant publication, plagiarism, and unethical research as retraction grounds. Version 3 keeps all of those and adds: image irregularities (manipulated or fabricated figures), fictitious data (invented results that were never collected), undisclosed use of artificial intelligence (where that use constitutes misrepresentation), and widespread systematic manipulation of the publication process, with paper mills named as the primary example. The authorship clause has also been sharpened. If authorship of the publication cannot be verified or there are serious concerns about accountability for the research, retraction is now explicitly on the table.

For most honest researchers, the image irregularities and authorship provisions are the most relevant. Image manipulation cases are now frequently caught by automated screening tools that journals use before and after publication, and the definition of manipulation in Version 3 broadly covers adjustments that change the scientific meaning of a figure rather than just improve its visual clarity. Journals are applying this retroactively to older papers when readers raise concerns, not only at submission.

Expressions of Concern: A Separate Framework

One of the most significant structural additions in the August 2025 release was a standalone guidance document on expressions of concern, separate from the retraction guidelines for the first time. Previously, COPE guidance on expressions of concern was embedded in the retraction document, which created confusion about when one was more appropriate than the other. The new document clarifies the distinction.

An expression of concern is not a retraction. It is a notice that a credible concern has been raised about a published paper and that an investigation is underway, but the outcome is not yet determined. The new guidance specifies that expressions of concern should be used when the journal has received credible information about a problem but cannot verify it yet, when an investigation is ongoing at an institution or funding body, or when the authors dispute the concern but the journal believes readers should be informed while the matter is resolved. The notice should describe the nature of the concern without prejudging the outcome.

From an author's perspective, receiving an expression of concern is serious but not final. The guidelines encourage journals to communicate with authors before posting one, to give them the opportunity to respond or to provide information that resolves the concern without a public notice. If you receive a journal email raising a concern about your paper, responding promptly and substantively is the single most effective thing you can do. An investigation that stalls because the authors are unresponsive tends to end in a worse outcome than one where authors engage and either correct the problem or establish that it was unfounded.

Batch Retractions and What Happens to Innocent Authors

The most genuinely new section of Version 3 covers batch retractions. This is the procedure that applies when a journal or publisher identifies a cluster of papers sharing the same underlying problem, usually paper mill origin, compromised peer review from a shared fake reviewer ring, or systematic data fabrication across a research group. Before 2025, journals handling these situations had to improvise. Some retracted papers one by one, each with a different notice. Others released bulk announcements. Version 3 now provides a defined framework.

Under the new guidance, when journals issue batch retractions, the retraction notice should explicitly state that systematic, coordinated, and widespread manipulation of the publication process has been identified and that the article is one of a group of papers affected by the same manipulation. This is a significant requirement. It means that a reader looking at your retraction notice should be able to tell whether you were retracted for an individual problem or caught up in a broader fraud that others orchestrated.

For innocent authors, Version 3 introduces an important procedural protection. In cases where an author was not involved in the fraud (for example, a genuine researcher whose name was added to a paper mill product without their knowledge, or a graduate student whose supervisor fabricated data), COPE now recommends that journal editors offer the author the opportunity for the article to be re-reviewed by independent reviewers before the retraction notice is published. This does not guarantee publication will be maintained, but it creates a formal checkpoint that was absent before.

The practical challenge is that batch retractions move fast. A publisher coordinating the retraction of 50 papers across multiple journals over a compressed timeframe may not individually contact every author with the same care they would apply to a single-paper case. If you ever receive a batch retraction communication, the guidance suggests requesting the re-review option explicitly if you believe your contributions to the paper were genuine and independent of whatever process the journal has flagged.

Retraction with Replacement and Retraction with Removal

Version 3 introduces two named forms of retraction that were not formalized in the previous guidelines, and the distinction matters for authors.

Retraction with replacement applies when the core findings of a paper remain valid but some portion of the original publication was flawed or missing in a way that a standard correction cannot adequately address. The retracted version is replaced by a corrected version that carries the same DOI or a linked DOI, with the retraction and replacement both documented in the notice. This option is appropriate when, for example, an analysis was run on a dataset that was later found to contain data entry errors, the errors have been corrected, and the conclusions remain substantially the same. It is less appropriate for data fabrication, where the underlying research record cannot be trusted.

Retraction with removal is reserved for cases where continuing to make the retracted paper publicly accessible could cause harm. The clearest examples are clinical case reports containing identifiable patient data published without consent, papers reporting dangerous drug doses or clinical procedures based on fabricated data, and papers involving legally protected identities. In most retractions, including those involving misconduct, the paper stays publicly accessible with a retraction watermark, because maintaining the record matters for anyone who cited it. Removal is rare and requires documented harm risk rather than editorial preference.

Timeliness: Why Journals No Longer Wait for Institutions

One of the most practically significant shifts in Version 3 is the explicit statement that delays in issuing retractions should not depend on institutional investigations or uncooperative authors. This has been a recurring problem in medical publishing for years. A journal identifies a problem, notifies the author's institution, and then waits months or years for the institution's investigation to conclude before acting. During that window, the paper continues to be cited. If it reports clinical findings, other researchers may build on results that the journal already has substantial reason to doubt.

Version 3 does not give journals the authority to act recklessly before an investigation concludes. But it does authorize them to act on the information they have, particularly when the problem is clear from the paper itself (obvious image duplication, for example), when the authors have been unresponsive for an unreasonable period, or when continued publication presents a specific risk to readers. The implication for authors is that being unresponsive to a journal inquiry does not protect you. It is more likely to accelerate the retraction than to delay it.

The new guidance also notes that journals should acknowledge third parties, such as post-publication reviewers, data integrity analysts, or the researchers who raised original concerns, in retraction notices when relevant and permitted. This reflects the growing role of services like Retraction Watch, PubPeer, and systematic post-publication review campaigns in surfacing problems. Papers published in the late 2010s and early 2020s that were never scrutinized during peer review are now being reviewed by organized groups of volunteers, and the new guidelines explicitly support crediting that work.

The Undisclosed AI Provision: What It Actually Covers

The provision treating undisclosed AI use as a potential retraction ground has received significant coverage, and some of that coverage has been more alarming than warranted. Version 3 does not make using AI in manuscript preparation a retraction offense. It makes undisclosed AI use a potential retraction offense when that use constitutes misrepresentation. The distinction is real and matters.

The framework specifies that journals should retract work when any form of misrepresentation, including the undisclosed use of artificial intelligence, is identified. The key word is misrepresentation. Using a language model to polish your discussion section and not mentioning it is, at most, a disclosure failure. The question of whether it rises to misrepresentation depends on how the journal interprets its own policy and whether the AI use affected the scientific content of the paper. Using AI to generate a literature review section that contained hallucinated citations, then submitting that section without disclosing the tool or checking the references, is much closer to misrepresentation.

The practical implication is that the COPE framework now explicitly supports journals that choose to retract for undisclosed AI use, when it is serious enough. This gives journals that were previously uncertain about their authority a clearer basis to act. Authors who have not yet established the habit of disclosing AI use in their manuscripts are now operating in a policy environment where the potential consequences have been formalized at the international guideline level. That is a meaningful change from 2019, when no AI provision existed at all.

What the Authorship Clause Means for Research Teams

Version 3's treatment of authorship disputes is more specific than the previous version and has particular relevance for clinical research teams that operate across institutions. The new text indicates that journals should retract work when authorship of the publication cannot be verified or there are serious concerns about accountability for the research. This is distinct from a routine authorship dispute, which journals typically handle through correspondence without retraction. It applies when the problem is that no one can actually verify who did what.

This matters most in two scenarios. The first is paper mill cases, where a submission agency may have purchased authorship slots and listed individuals who had no involvement in the research. The second is large collaborative studies where, after publication, no named author can produce the raw data, confirm that the study was actually conducted, or demonstrate meaningful engagement with the manuscript. Both scenarios have become more common as investigators face pressure to publish in high volumes and some research groups outsource aspects of manuscript preparation to third parties.

The CRediT taxonomy of author contributions, now mandatory at most major publishers, exists partly to address this problem before publication. A detailed CRediT statement that names who collected the data, who ran the analysis, and who verified the results makes it much harder for a journal to later claim that authorship accountability is unclear. If your institution or research group has not yet adopted CRediT as standard practice, the strengthened COPE authorship provisions are a practical reason to start.

What All of This Means Before You Submit

Retraction policy tends to feel like something that happens to other people. For most researchers, that is true. The majority of published medical papers are never subject to any integrity investigation, and the number of retractions in any given year, while growing, remains a small fraction of total output. But the COPE framework changes what is possible when problems are found, and knowing how that framework works gives you more control over how you position yourself.

On figures: if you are processing microscopy images, gel photographs, or any clinical imaging, document every processing step before submission. Keep the raw files alongside the processed ones. If you are asked to show your original data post-publication, being able to produce it immediately is far better than spending weeks locating it or discovering that a junior researcher deleted the originals. The image integrity provision in Version 3 does not distinguish between deliberate manipulation and careless processing that changed the result. The journal will investigate the effect, not your intention.

On AI: the undisclosed AI provision formalizes what most major journals were already signaling. If you used a generative tool during manuscript preparation, say so. The disclosure does not need to be elaborate. Name the tool, name the section it touched, confirm that a human reviewed and takes responsibility for the output. A sentence in your methods or acknowledgements is harder to argue with than a silence that a reviewer or reader later notices.

On authorship: before you submit, every named author should be able to answer the question: what exactly did you do, and can you show me? That is not a hostile question. It is the CRediT taxonomy applied to your own paper. If someone is listed as an author who does not know the answer, that is worth resolving before the paper is published rather than after a problem surfaces and the journal asks the same question in a less comfortable context.

The core purpose of the COPE guidelines, explicitly stated in Version 3, is to correct the literature rather than to punish individuals. That framing is genuinely useful. Most integrity problems that reach journals are not cases of deliberate fraud by the named authors. They are cases where someone cut a corner, did not verify something they should have verified, or was unaware that a process used in their research group was considered problematic. The authors who navigate these situations best are the ones who engage transparently from the first contact and provide documentation rather than explanations.

Further Reading