CASRAI Dictionary

Tag: COPE Guidelines

  • Elsevier’s Research Integrity Screening Process

    Elsevier screens research submissions for integrity issues through a layered pipeline: automated tools such as Check Integrity and Crossref Similarity Check flag plagiarism, duplicate submissions and image anomalies at intake, specialist Research Integrity and Publishing Ethics (RIPE) analysts investigate confirmed concerns, and outcomes range from correction through expression of concern to full retraction, following guidelines set by the Committee on Publication Ethics (COPE).

    Research integrity screening is the set of technical checks and human review stages a publisher applies to a manuscript, before and after publication, to detect fabrication, falsification, plagiarism, undisclosed image manipulation and paper-mill activity. At Elsevier, that pipeline runs continuously from the moment a manuscript is submitted to the point, if necessary, of retraction.

    How Elsevier’s research-integrity pipeline works, from submission to retraction

    Elsevier operates one of the largest editorial screening operations in scholarly publishing. In 2025, the publisher received 4.2 million manuscript submissions across roughly 3,000 journals and published 795,000 after validation and peer review, according to Elsevier’s own account of its editorial process. Elsevier states that its published output accounts for over 18% of global research output and 29% of citations — a scale that shapes why it has invested heavily in both automated screening and dedicated integrity staff rather than relying on peer review alone.

    The pipeline runs across four broad stages, each with a different primary tool or team responsible for catching a different class of problem.

    Stage Primary tool or team Typical trigger
    Submission intake Check Integrity screening tool; Crossref Similarity Check (iThenticate) Text overlap, duplicate manuscript, unauthorised authorship change
    Peer review Editors, external reviewers, RIPE analysts Implausible data, reviewer-flagged inconsistency, suspicious image reuse
    Post-publication monitoring Research Integrity and Publishing Ethics (RIPE) team Reader or whistleblower reports, cross-journal pattern analysis
    Enforcement Editors-in-chief, following COPE-guided process Confirmed fabrication, falsification or plagiarism

    What does Elsevier screen for at the point of submission?

    Every manuscript submitted to an Elsevier journal is routed through automated checks before an editor sees it. Check Integrity, Elsevier’s proprietary screening tool, had been expanded across more than 2,000 journals as of March 2026, according to trade press coverage in Research Information. The tool automatically reviews submissions for red flags — including unauthorised authorship changes, undisclosed conflicts of interest and signs of duplicate or template-like submission — and routes anything flagged to specialist integrity analysts, freeing editors to focus on scientific merit.

    Plagiarism screening runs in parallel through Crossref Similarity Check, powered by iThenticate, which compares submitted text against a large index of published articles and web content. There is no fixed similarity percentage that automatically triggers rejection; editors interpret each report to distinguish appropriate citation from genuine textual misconduct.

    Paper-mill detection layers on top of these checks. Integrity analysts look for patterns that recur across industrialised fraud, including:

    • Formulaic, template-like titles or methods sections
    • Unusual or inconsistent author affiliations and contact details
    • Data or experimental descriptions that do not match the stated methodology
    • Systematic image reuse across ostensibly unrelated papers
    • Irregular peer-review patterns, such as reviewer suggestions tied to the same small pool of contacts

    How does Elsevier detect image manipulation and data-integrity problems?

    Image screening combines editorial guidelines with a mix of manual and software-assisted checks. Elsevier’s policy permits minor adjustments to brightness, contrast or colour balance only where they do not obscure or eliminate information present in the original image; the use of generative AI to create or alter a figure is prohibited outright. Where manipulation is suspected, editors can apply forensic image-analysis tools of the kind recommended by the US Office of Research Integrity, and will typically request the original, unprocessed image files directly from the authors.

    Elsevier has also published on the scale of automated flagging behind these checks. At the 8th World Conference on Research Integrity in 2024, Elsevier data scientist Yuri Kashnitsky presented on large-scale flagging of integrity misconduct across the publisher’s portfolio, noting that all system-generated findings are manually checked and confirmed by investigators before any corrective action is suggested to editors — underscoring that software narrows the search space, but a human analyst still makes the determination.

    Who investigates confirmed misconduct, and what enforcement follows?

    Once a concern is substantiated, Elsevier’s in-house Research Integrity and Publishing Ethics (RIPE) team leads the investigation, working with journal editors and, where warranted, the authors’ institutions. Elsevier states that it follows retraction guidelines developed by COPE, and confirmed problems resolve into one of three outcomes: a correction or erratum for errors that do not undermine the paper’s conclusions, an expression of concern where the investigation is inconclusive but doubts remain, or a retraction where the findings are no longer considered reliable.

    A recent case shows this enforcement ladder operating at scale. In a statement updated in May 2026, Elsevier disclosed that a comprehensive, multi-year audit of the journal Heliyon — using Check Integrity screening combined with manual review by RIPE analysts — had produced approximately 1,100 corrections to the scientific record, affecting around 3% of everything the journal had published across 12 years. Those 1,100 actions spanned corrections, expressions of concern and retractions; impacted authors were notified and given the chance to respond before editors made a final determination. Following the audit, Web of Science removed an indexing hold it had placed on Heliyon, and Elsevier said it was applying lessons from the case to workflows across its wider journal portfolio.

    Common questions about Elsevier’s integrity screening

    Does Elsevier use iThenticate for plagiarism screening?

    Yes. Elsevier’s journals route submitted manuscripts through Crossref Similarity Check, which is powered by iThenticate, comparing text against a large index of published articles and web content. Editors, not the software alone, judge whether flagged overlap reflects proper citation or genuine plagiarism before any editorial decision is made.

    Who investigates allegations of research misconduct at Elsevier?

    Elsevier’s in-house Research Integrity and Publishing Ethics (RIPE) team investigates confirmed concerns, working alongside journal editors and, where relevant, the authors’ institutions. Investigations follow COPE guidelines and typically involve requesting raw underlying data before any corrective action is taken.

    What is considered the most serious form of research misconduct?

    Fabrication and falsification of data are generally treated as the most serious forms of misconduct, alongside plagiarism, because they directly corrupt the reliability of the published record. Elsevier’s policies place these above lesser breaches such as citation gaming or unresolved authorship disputes.

    What happens after a research-integrity investigation confirms a problem?

    Confirmed issues lead to one of three outcomes: a correction for errors that do not undermine the findings, an expression of concern where evidence is inconclusive, or a retraction where the results are no longer considered reliable. All three are published and linked to the original article, per COPE guidance.

    What this means for institutions, authors and integrity offices

    For research administrators, the Heliyon case is a reminder that publisher-side screening is a complement to institutional processes, not a substitute for them. When a journal’s RIPE team contacts an institution about a flagged submission or published paper, that request typically triggers — and depends on — the institution’s own research-integrity office and record-keeping, an area covered in more detail in CASRAI’s research administration resources and its wider research-integrity dictionary entries. Authors, in turn, should expect to be asked for raw, unprocessed data or images at any stage, including years after publication, and should retain those records accordingly.

    Elsevier is not acting alone: it collaborates with other publishers through the STM Integrity Hub to detect duplicate submissions across the wider industry, reflecting a broader shift toward cross-publisher, not just single-journal, integrity infrastructure. As automated screening tools mature, the balance is likely to keep shifting toward earlier detection at submission — but the Heliyon audit shows that human RIPE analysts, not algorithms, remain the ones who make the final call on correction, expression of concern or retraction.

  • Editorial Expression of Concern vs Retraction

    An editorial expression of concern is a notice a journal publishes to flag credible but unresolved doubts about a paper’s reliability, without retracting it. Editors use it instead of a retraction when the evidence is inconclusive, an institutional investigation is ongoing, or a fair resolution will take months rather than weeks.

    An editorial expression of concern (sometimes abbreviated EEoC) is defined by the Committee on Publication Ethics (COPE) as a journal notice about potential misconduct or unreliable findings, issued when the available evidence does not yet meet the threshold for a correction or a retraction. It sits deliberately between silence and withdrawal — a middle mechanism the scholarly record uses to signal risk without pre-judging guilt.

    What is an editorial expression of concern?

    An editorial expression of concern is a public, freely accessible notice — linked bidirectionally to the article it concerns — that tells readers a paper’s integrity is in question. Wikipedia’s entry on the topic describes it simply as “a notice issued by a publisher against a particular publication, warning that it may contain errors or be otherwise untrustworthy,” citing Morris, Barnas, LaFrenier and Reich’s Handbook of Journal Publishing (Cambridge University Press, 2013).

    Crucially, an expression of concern does not itself amend the scientific record. The original article remains published, unaltered, alongside the notice. It is a flag, not a verdict — and that distinction is precisely why journals reach for it before they reach for a retraction.

    Expression of concern vs retraction: what’s the difference?

    A retraction is a formal withdrawal: editors have concluded, on clear grounds, that a paper’s findings are unreliable because of fabrication, falsification, plagiarism, an invalidating honest error, or an unresolved ethical breach. COPE’s Retraction Guidelines (Version 3, 2025) set out the timing, content and evidentiary bar for that step. A retraction changes how the article is labelled in perpetuity; it does not usually remove the text, but it marks it as withdrawn from the reliable literature.

    An expression of concern makes no such finding. It is an interim or, occasionally, a terminal notice used precisely because the evidence does not yet support — or may never support — a definitive retraction decision. The International Committee of Medical Journal Editors (ICMJE), in its December 2019 recommendations on “Scientific Misconduct, Expressions of Concern, and Retraction,” acknowledges that a publisher may issue an expression of concern while a misconduct investigation is ongoing, and pending its outcome, rather than waiting in silence or retracting prematurely.

    Feature Expression of concern Retraction
    Evidence threshold Inconclusive or investigation ongoing Clear grounds established
    Effect on the article Article stands, flagged with a linked notice Article marked withdrawn from the record
    Typical trigger Credible allegation, pending institutional inquiry Confirmed fabrication, falsification, plagiarism or invalidating error
    Finality Interim — or occasionally the final outcome if no further update is expected Final

    When does COPE say editors should issue one?

    COPE’s dedicated guideline on expressions of concern gives editors concrete criteria. It states that editors should consider issuing an expression of concern if significant and credible concerns have been raised but the evidence is unclear as to whether the work, or parts of it, are potentially unreliable — whether from error, incorrect analysis, or research-integrity concerns affecting the main findings.

    Other qualifying scenarios include:

    • An institutional, funding, or other formal oversight investigation is ongoing and may lead to corrections based on the work’s reliability.
    • Authors have been asked for additional information to address concerns, and it is not immediately available.
    • There is an unresolved breach of journal or publisher policy — for example, data that were available at publication but later withheld.
    • A resolution is not expected for some time, typically several months.

    COPE is equally clear about when an expression of concern is not appropriate: if the editor can quickly reach a decision, if resolution is only weeks away, if the main findings remain reliable, or if the sole concern relates to authorship rather than the underlying findings. Publishing a notice that will be superseded within weeks risks confusing readers rather than informing them.

    Corrigendum vs erratum vs expression of concern: where each notice fits

    Expressions of concern sit alongside — but are distinct from — the more routine correction notices journals issue. In standard publishing practice, a corrigendum is a correction initiated by the authors themselves to fix an error they introduced (a wrong affiliation, a miscalculated value, an omitted co-author), while an erratum corrects a mistake introduced by the publisher during production or typesetting — the classic corrigendum vs erratum distinction. Neither implies a reliability concern about the underlying findings; both simply amend the published record.

    The National Information Standards Organization’s 2024 recommended practice, NISO RP-45-2024, “Communication of Retractions, Removals, and Expressions of Concern (CREC),” formalises how publishers and indexers should structure and disseminate these various post-publication notices so that the scholarly record — and the systems that index it — stay consistent across platforms.

    Notice type Who initiates it What it signals Effect on the record
    Corrigendum Authors Author-side error in an otherwise sound paper Text amended; original findings stand
    Erratum Publisher Production or typesetting error Text amended; original findings stand
    Expression of concern Editors/publisher Credible, unresolved doubt about reliability Article stands, flagged pending outcome
    Retraction Editors/publisher Confirmed unreliability or misconduct Article marked withdrawn

    Related questionable research practices — undisclosed image manipulation, salami-slicing, or selective reporting — often surface first through post-publication scrutiny on platforms such as PubPeer, which can prompt an editor to move from silence toward one of these four notice types.

    Frequently asked questions

    What is an editorial expression of concern?

    An editorial expression of concern is a notice issued by editors or a publisher to draw attention to potential problems in a published paper, without itself constituting a retraction or a correction. It flags unresolved doubt while the article remains part of the published record.

    What is an example of when an expression of concern applies?

    COPE gives the example of an editor receiving inconclusive evidence of misconduct, or learning that findings appear unreliable but the authors’ institution declines to investigate. In both cases, the concern is credible but not yet provable, so a flag — not a withdrawal — is the appropriate response.

    What is the difference between an expression of concern and a retraction?

    An expression of concern is a provisional flag issued when evidence is incomplete; a retraction is a final, formal withdrawal issued once editors have established clear grounds — fabrication, falsification, plagiarism, or an invalidating error. One pauses judgement; the other delivers it.

    Is Retraction Watch a credible source for tracking these notices?

    Retraction Watch is widely cited by researchers, journalists and integrity officers as a tracking resource and maintains a large public database of retractions. It is a secondary aggregator, not a standards body — for authoritative process guidance, COPE, ICMJE and NISO remain the primary reference sources.

    Implications for institutions, publishers and researchers

    For research administrators and institutional integrity officers, an expression of concern on a faculty member’s paper is not proof of misconduct — but it is a signal that warrants tracking, particularly where funding, promotion, or REF-style assessment exercises depend on the work’s standing. Institutions should distinguish, in their own case-management records, between papers carrying a corrigendum or erratum (routine) and those carrying an expression of concern (an active, unresolved integrity question).

    For publishers and journal editors, COPE’s criteria function as a due-process safeguard: they prevent both premature retraction, which can unfairly damage careers, and prolonged silence, which leaves readers citing potentially unreliable findings unwarned. NISO RP-45-2024’s structured communication requirements push this further, aiming to make expressions of concern discoverable wherever an article is indexed, not just on the publisher’s own site.

    As post-publication scrutiny — via PubPeer, institutional audits, and journal-side data checks — continues to intensify, expressions of concern are likely to become a more visible, more standardised fixture of the published record, sitting permanently between the routine correction and the definitive retraction.

  • Corrigendum vs Erratum: A Decision Framework

    A corrigendum corrects an error the authors introduced before publication — flawed data, a wrong affiliation, a miscalculation — while an erratum corrects an error the publisher introduced during production, such as a typesetting mistake or a misplaced figure. A retraction withdraws the paper entirely because its findings can no longer be trusted. Which notice an editor issues depends on who caused the error and whether it invalidates the paper’s conclusions.

    A corrigendum vs erratum decision is one of the most common — and most frequently mishandled — calls a journal editor or research integrity office makes. Get it wrong and a paper that needed only a correction ends up stigmatised as retracted, or worse, a paper with compromised findings survives under a cosmetic “correction” label. This explainer sets out a COPE-aligned decision framework for distinguishing corrigenda, errata, and retractions, and for knowing when an interim Expression of Concern is the right holding step.

    Definition: a corrigendum is a post-publication notice correcting an error the authors introduced into an already-published article, distinct from an erratum, which corrects an error the journal or publisher introduced during production.

    Contents

    What Is the Difference Between a Corrigendum and an Erratum?

    The Committee on Publication Ethics (COPE) states the working distinction plainly: “the term erratum usually refers to a production error, caused by the journal,” while “the term corrigendum (or correction) usually refers to an author error.” An erratum fixes something the journal’s own production process broke — a dropped decimal point during typesetting, a caption swapped between two figures, an author’s name misspelled by a copyeditor. A corrigendum fixes something wrong in the manuscript the authors submitted — an incorrect data table, a miscalculated statistic, an omitted funding disclosure, a wrong affiliation.

    Both notices share one condition: the underlying conclusions of the paper must remain valid. Neither is a vehicle for revising results, adding new data, or reversing a finding — that requires a different mechanism entirely.

    Not every publisher enforces this line consistently. COPE itself notes that “many journals do not distinguish between corrigendum or erratum but the note can explain if it is the publisher’s or authors’ error,” and some use “correction” as a catch-all. The US National Library of Medicine, which indexes PubMed, goes further still: it classifies both corrigenda and errata under a single “Published Erratum” tag for cataloguing purposes, regardless of who caused the error. Editors should not assume a downstream index preserves the author/publisher distinction they carefully applied upstream.

    Notice type Who caused the error Typical trigger Effect on the record
    Erratum Publisher/production Typesetting, misplaced figure, production-introduced typo Article stands; notice linked bidirectionally
    Corrigendum Author(s) Data entry error, wrong affiliation, miscalculation Article stands; notice linked bidirectionally
    Addendum Author(s) New, non-conflicting supplementary information Article stands; addendum appended
    Expression of Concern Unresolved Serious concern raised, investigation incomplete Article flagged pending outcome
    Retraction Author(s), publisher, or both Unreliable findings, misconduct, plagiarism, duplicate publication Article formally withdrawn; original remains visible with notice

    When Does an Error Require a Retraction Instead of a Correction?

    A retraction is warranted the moment an error stops being cosmetic and starts undermining the paper’s conclusions. COPE’s retraction guidance and the ICMJE Recommendations converge on the same threshold: if the reported findings can no longer be relied upon — whether from honest experimental error, data fabrication, image manipulation, or a flawed method that invalidates the results — a correction notice is insufficient and the paper must be retracted.

    COPE’s published criteria for retraction include clear evidence of unreliable findings, whether from major error or research misconduct; plagiarism or duplicate/redundant publication without proper attribution; and research that failed to meet ethical requirements for human or animal subjects. A single miscoded statistical test that changes a paper’s headline conclusion crosses this line; a mislabelled axis on a supplementary figure does not.

    This is a binary test editors should apply explicitly rather than by instinct: does the error change what a reader would conclude from the paper? If yes, retraction (or, where only part of the paper is compromised, a carefully scoped partial retraction) is the correct instrument — not a corrigendum dressed up to avoid reputational fallout.

    A Step-by-Step Decision Framework for Editors

    Use this sequence to route an identified error to the correct notice type:

    1. Does the error affect the validity, reliability, or interpretation of the results? If no, proceed to step 2. If yes, skip to step 4.
    2. Who introduced the error — the authors before submission, or the publisher during production? Author-origin errors that don’t affect conclusions become a corrigendum; publisher-origin errors become an erratum.
    3. Is the correction additive rather than corrective — new information that doesn’t conflict with the published record? Route to an addendum instead.
    4. Is there evidence of misconduct, fabrication, falsification, plagiarism, or duplicate publication, or does the error invalidate the paper’s central conclusions? If yes, this is retraction territory.
    5. Is an investigation still open but the concern serious enough that readers need an immediate warning? Issue an Expression of Concern as an interim notice while the inquiry proceeds.
    6. Publish the notice with bidirectional linking to the original article’s DOI, so indexers, Crossref, and reference managers can propagate the correction status automatically.

    Every step should be documented in the journal’s editorial record even when the final notice seems self-evident — COPE explicitly cautions that authors must be informed before any correction or retraction notice is published, and that agreement on wording, while ideal, “is not a requirement.”

    Expressions of Concern, Partial Retractions, and Addenda

    Three intermediate instruments sit between “leave it alone” and “retract in full,” and editors under-use all three.

    • Expression of Concern: issued when an investigation into a paper’s integrity is under way but unresolved, and the concern is serious enough that readers need warning before the inquiry concludes. It is explicitly a holding notice, not a verdict.
    • Partial retraction: withdraws only the compromised portion of a paper — a single figure, dataset, or claim — while leaving the remainder of the record intact. COPE guidance treats this instrument cautiously, since a partially retracted paper can be difficult for readers to interpret correctly; it should be reserved for cases where the surviving content is genuinely independent of the flawed portion.
    • Addendum: adds new, non-conflicting information to a published paper — for example, a dataset the authors did not originally deposit. It corrects nothing; it supplements.

    Since 2023, Crossref has stewarded the Retraction Watch Database and integrated its records into Crossref metadata, making retraction status machine-discoverable directly from an article’s DOI rather than requiring a manual search. This matters operationally: a research office checking whether a cited paper has been retracted, corrected, or flagged with an Expression of Concern can now query that status programmatically instead of relying on the journal’s own website staying current.

    Redaction vs Retraction: Why the Terms Get Confused

    “Redaction” and “retraction” are frequently conflated in search queries and in casual use, but they describe unrelated actions. Redaction means removing or obscuring specific sensitive content from a document — a patient identifier, classified information, a legally privileged passage — while leaving the rest of the document intact and published. Retraction means formally withdrawing an entire scholarly article from the reliable literature because its findings can no longer be trusted.

    A research office redacts personal data from a dataset before deposit; it does not redact a published paper. If the concern is that a paper’s findings are unreliable, the applicable mechanism is a correction (corrigendum/erratum) or a retraction — never a “redaction” of the article itself. Editors and compliance staff should treat any external query using “redact” in connection with a published paper as a signal to clarify which of the five instruments above the requester actually means.

    Common Questions About Corrections and Retractions

    What Is the Difference Between Addendum and Corrigendum?

    An addendum adds new, non-conflicting information to a published paper, such as a supplementary dataset the authors omitted at submission. A corrigendum corrects an existing error the authors introduced — incorrect data, a wrong affiliation, a miscalculation. Addenda supplement the record; corrigenda fix it.

    What Is the Purpose of a Corrigendum?

    A corrigendum exists to correct an author-introduced error that appears in the body of a published article — flawed data, a misreported figure, or an incorrect reference — without withdrawing the paper. It preserves the article’s validity while making the published record accurate and permanently linked to the original.

    What Is the Difference Between Errata and Correction?

    “Correction” is the umbrella term covering any post-publication fix that doesn’t invalidate a paper’s conclusions. “Errata” traditionally denotes production-side corrections specifically, but the US National Library of Medicine indexes all such fixes — author or publisher caused — under a single “Published Erratum” category, regardless of origin.

    What Does an Erratum Mean?

    An erratum is a formal notice correcting an error the publisher or journal introduced during production — typesetting, a misplaced figure, a formatting mistake — rather than an error present in the authors’ original submission. It confirms the underlying research and conclusions remain valid.

    Implications for Research Offices

    For institutional research offices and research integrity staff, the practical stakes of this distinction are higher than the terminology suggests. Grant reporting, tenure and promotion dossiers, and REF/assessment-exercise submissions all treat “retracted” very differently from “corrected” — a retraction can trigger funder inquiries and institutional misconduct review, while a corrigendum typically does not. Misclassifying a serious, conclusion-altering error as a corrigendum to avoid that scrutiny is itself a publication-ethics failure, not a shortcut.

    Research administrators should build a standing check into any publication-tracking workflow: query DOIs against Crossref’s retraction metadata at the point of citation-count reporting or dossier compilation, not only at initial publication, since correction and retraction notices can post months or years after the original article.

    The Bottom Line

    The corrigendum/erratum boundary is a question of who caused the error; the correction/retraction boundary is a question of whether the paper’s conclusions still hold. Editors who apply both tests explicitly — and who reach for an Expression of Concern when an investigation is still open — protect the scholarly record more effectively than those who default to whichever notice feels least reputationally costly. As Crossref’s retraction metadata becomes a standard citation-checking layer, the cost of misclassification is no longer just reputational; it is now machine-visible across the scholarly infrastructure that research offices, funders, and publishers all query.

    For related terminology, consult the CASRAI Dictionary of research-administration terms, and see how correction and retraction tracking fits into wider institutional workflows on the research administration hub.

  • Research Misconduct Statistics: What Springer Nature’s 2025 Retraction Data Reveal

    Springer Nature’s 2025 research-integrity disclosure landed with a number that cuts against the usual narrative: 1,462 retractions across its portfolio, roughly half the 2,923 logged in 2024. Read at face value, that looks like progress. Read against the underlying research misconduct statistics, it looks more like a legacy backlog being worked through than a crisis being resolved — 57% of 2025’s retractions (833 articles) were for papers published before January 2024, meaning the majority of this year’s corrections trace back to older, previously accumulated problems rather than newly discovered misconduct. For institutions, publishers and funders, that distinction changes the risk calculus considerably.

    Springer Nature’s 2025 Retraction Snapshot

    Springer Nature published these figures on its public research-integrity page, alongside its 2024 comparator, offering a rare year-on-year, publisher-disclosed dataset rather than a third-party estimate.

    Metric 2024 2025
    Total retractions 2,923 1,462
    Share for pre-cut-off papers 61.5% (1,797) — before Jan 2023 57% (833) — before Jan 2024
    Share for post-cut-off papers 38.5% (1,126) — after Jan 2023 43% (628) — after Jan 2024
    Post-cut-off retractions that were open access 41% ~21%
    Articles published that year 482,000+ 539,000
    Submissions received 2.3 million 3.1 million

    Set against roughly 539,000 primary research articles published in 2025, the 1,462 retractions represent under 0.3% of that year’s output — consistent with long-standing academic estimates that outright fabrication or falsification affects a small minority of the literature, even as absolute retraction counts have climbed industry-wide over the past decade.

    Backlog-Clearing or a Rising Tide?

    Two things are true at once. Springer Nature’s own retraction count fell by roughly half between 2024 and 2025. But the proportion attributable to legacy, pre-cut-off papers barely moved — 61.5% in 2024, still 57% in 2025 — which means well over half of each year’s retraction activity is publishers working backwards through their archive, not reacting to current misconduct.

    That pattern sits inside a wider industry trend. Nature reported that more than 10,000 papers were retracted across all publishers in 2023 — an all-time record at the time, driven substantially by mass clean-ups at journals compromised by paper mills. Springer Nature’s 2025 dip suggests one large publisher has made a dent in its own backlog, not that the sector-wide correction cycle has ended.

    • Legacy-paper retractions remained the majority share in both 2024 and 2025.
    • The open-access share of post-cut-off retractions nearly halved year on year (41% to ~21%), a data point worth monitoring rather than celebrating in isolation.
    • Springer Nature’s book-integrity investigations followed a similar arc: 124 in 2022, 207 in 2023, 217 in 2024, 210 in 2025, and 81 already by mid-April 2026 — prompting the publisher to introduce editorial expressions of concern for books in 2026.

    Root Causes: Paper Mills, Flawed Datasets and Peer-Review Fraud

    Springer Nature attributes its retractions to a recurring set of causes, echoed across the wider Retraction Watch record: data fabrication or falsification, plagiarism and duplicate publication, compromised or fraudulent peer review, unresolved authorship or consent issues, and the systematic activity of paper mills — commercial operations selling fabricated manuscripts or authorship slots.

    A live 2025 case illustrates how these risks travel across publishers. Springer Nature began retracting or removing 38 papers, conference proceedings and book chapters that trained neural networks on a dataset of children’s facial images scraped from autism-related websites without verifiable consent or diagnostic confirmation. Wiley had separately retracted two papers using the same dataset in 2023, and researchers identified at least 90 citing publications across the industry, with IEEE confirming an active investigation. One flawed dataset, multiple publishers, years of downstream exposure — a pattern institutional risk officers should recognise.

    What percentage of scientific papers are retracted?

    At Springer Nature, 1,462 retractions against roughly 539,000 articles published in 2025 equals under 0.3% of that year’s output. Broader academic surveys estimate outright misconduct — fabrication, falsification or plagiarism — affects between 0.3% and 4.9% of published research, depending on definition, discipline and detection method.

    Why are research papers retracted?

    Papers are retracted when the integrity of published work is substantially undermined — through data fabrication, plagiarism, compromised peer review, undisclosed authorship or consent problems, or paper-mill involvement. Retractions can also follow honest error, and are sometimes initiated by authors themselves once a flaw is confirmed, per COPE guidance.

    What is the difference between a retraction and an editorial expression of concern?

    An editorial expression of concern is an interim, indexed notice flagging serious unresolved concerns while an investigation continues. A retraction is the final editorial decision, made once integrity is confirmed as substantially compromised, following Committee on Publication Ethics (COPE) best-practice guidelines.

    What This Means for Institutional Risk Exposure

    Because well over half of each year’s retractions attach to papers published one, two or more years earlier, institutions cannot treat retraction risk as a current-cycle problem. Grant reports, tenure and promotion files, systematic reviews, and REF-style assessment submissions can all cite work that is retracted retroactively, with reputational and funding consequences that surface long after the original publication date.

    That is precisely why structured, per-contributor attribution matters. CASRAI originated the CRediT contributor role taxonomy in 2014; the standard is now stewarded by NISO as ANSI/NISO Z39.104-2022. Where a CRediT contributor role statement clearly separates who ran the analysis, who supplied data, and who supervised the work, institutions and journals can isolate accountability far more precisely than a flat author byline allows — a distinction that becomes material the moment a co-authored paper is flagged. Research administration offices should treat this as core infrastructure, not paperwork: clear authorship documentation shortens investigation timelines and protects contributors who had no role in the disputed element of a paper.

    Publishers are also expanding scrutiny beyond journal articles. Springer Nature’s move to issue expressions of concern for books, after growing its book-related integrity probes from 124 to over 200 a year, signals that monograph and chapter output — historically under-scrutinised — now carries comparable institutional exposure to journal articles.

    Looking Ahead: How Institutions Should Respond

    Springer Nature’s figures update twice yearly, and the publisher has signalled that legacy-paper clean-up is an ongoing commitment rather than a one-off exercise — meaning the majority-legacy retraction pattern is likely to persist for several more reporting cycles. For research administration teams, that argues for a shift from reactive incident response to standing audit practice.

    • Audit legacy institutional outputs against publisher retraction and expression-of-concern notices, not just current submissions.
    • Require structured CRediT-style contributor statements on new submissions to enable faster, fairer accountability if a paper is later flagged.
    • Track publisher-level transparency pages (Springer Nature, and equivalents at other major publishers) alongside COPE guidance and the Retraction Watch database as standing monitoring sources.
    • Extend integrity oversight to books and monographs, not only journal articles, given publishers’ expanding scrutiny in this area.

    The headline number fell in 2025. The underlying research misconduct statistics say the correction cycle for legacy scholarship is far from finished — and institutions that plan accordingly, rather than reading a single year-on-year dip as resolution, will be better placed for whatever the next reporting cycle reveals.

  • Retraction in Academic Publishing: A Terminology Guide for Editors and Research Offices

    Editors, research-integrity officers and authors routinely use “retraction”, “correction”, “expression of concern” and “redaction” as if they were interchangeable. They are not. Retraction in academic publishing is a formal, permanent withdrawal of a paper’s standing, reserved for findings that can no longer be trusted — a different remedy, with a different evidence threshold, from a correction, an expression of concern, or a redaction. Conflating the terms slows investigations and can misstate a case’s severity to funders, tenure committees, and the public record.

    The four publishing remedies at a glance

    The Committee on Publication Ethics (COPE), whose retraction guidelines were updated to version 3 in August 2025, and the International Committee of Medical Journal Editors (ICMJE) both treat these as distinct editorial tools, not synonyms. The table below sets out the working distinctions research offices and editorial staff need.

    Remedy What it means Typical trigger Who issues it Effect on the original article
    Retraction Formal, permanent withdrawal from the reliable literature Fabrication, falsification, plagiarism, unethical research, compromised peer review Editor, sometimes jointly with the publisher Article stays online, clearly watermarked “RETRACTED”
    Correction (corrigendum/erratum) Fixes a specific, non-fatal error Author error (corrigendum) or production error (erratum) Authors (corrigendum) or journal (erratum) Article stands; correction notice is linked to it
    Expression of concern Interim public notice pending an unresolved inquiry Inconclusive evidence or an ongoing institutional investigation Editor Article stands, flagged as under review
    Redaction Removal or masking of specific sensitive content only Legal, privacy, or confidentiality requirement (e.g. identifiable patient data) Publisher, usually on legal or data-protection advice Only the redacted portion is withheld; the rest of the record stands

    Retraction: when findings cannot be trusted

    Per COPE’s 2025 retraction guidelines, “the purpose of retraction is to correct the literature and ensure its integrity, not to punish the authors.” A retraction disavows the paper’s conclusions; it does not usually remove the text itself, which stays accessible but permanently marked as unreliable.

    COPE and ICMJE both point to a similar set of grounds for retraction:

    • Unreliable findings — from honest error (miscalculation, flawed methodology) or from misconduct (fabricated or falsified data)
    • Plagiarism — appropriating another party’s words, data, or ideas without credit
    • Redundant or duplicate publication — the same findings published elsewhere without cross-reference or permission
    • Unethical research — studies that breached human- or animal-subject ethics requirements
    • Compromised peer review or undisclosed conflicts of interest that could have biased the editorial decision

    A retraction can be initiated by authors, editors, or the publisher, but COPE is explicit that the editor holds final decision authority, consistent with editorial independence — the publisher’s role is to support the investigation and help issue the notice, not to make the call.

    Correction, corrigendum and erratum: fixing the record without withdrawing it

    A correction is the appropriate remedy when an error is real but does not undermine the paper’s overall conclusions — a mislabelled figure, an incorrect affiliation, a transposed digit in a table. Two related terms are often used loosely but have a real distinction:

    • Corrigendum — a correction of an error introduced by the authors themselves
    • Erratum — a correction of an error introduced by the journal during production

    Both are published as a linked notice attached to the original article, which otherwise remains part of the reliable record. A correction is not a lesser form of retraction — it is a separate remedy for a separate class of problem, and treating minor corrections as reputational events discourages the self-correction that COPE and ICMJE actively encourage.

    Expression of concern: the interim signal

    An expression of concern (EOC) is not a verdict. COPE’s guidance describes it as the appropriate step when an editor is uncertain about a publication’s reliability because of insufficient information, delays in institutional response, or an investigation that will not conclude quickly. Rather than wait — and risk the paper being cited or acted on in the meantime — the editor publishes a notice flagging the concern while the inquiry continues.

    An EOC typically resolves in one of three ways: retraction, correction, or a formal confirmation that the concerns did not hold up. Editorial and research-integrity teams should track EOCs as open cases, not closed ones, and revisit them on a defined schedule rather than leaving them unresolved indefinitely.

    Redaction vs retraction: a different kind of removal

    This is where terminology confusion is most common — and most consequential. Redaction is not a recognised category within the COPE/ICMJE retraction-correction-EOC taxonomy. It is a records-management and legal term for the selective removal or masking of specific sensitive content — identifiable patient information, confidential commercial data, material under a court order — while the rest of the document remains intact and in force.

    Retraction, by contrast, withdraws the paper’s standing as a whole. A journal might redact one identifying detail from a case report to comply with data-protection law without touching the paper’s scientific conclusions; that is not equivalent to, and should never be reported internally as, a retraction. Research offices logging cases for funder reporting should keep these as separate fields — collapsing them into one “removed” category misrepresents both the scale and the cause of the action.

    Answer-first questions editors ask

    What does it mean if a publication is retracted?

    A retracted publication has been formally withdrawn from the reliable scholarly record by its editor, usually because of unreliable data, misconduct, or a serious ethical breach. The article typically remains online, watermarked “RETRACTED”, so the record stays transparent rather than being erased.

    What is the purpose of retractions in academic publishing?

    Retraction exists to correct the literature and protect its integrity, not to punish authors, per COPE’s own guidelines. It warns future readers, citing authors, and clinicians not to rely on the paper’s findings or conclusions, limiting downstream harm from erroneous or fraudulent results.

    Do retracted studies still get cited?

    Yes — research tracked via PubMed Central and the Retraction Watch database shows retracted papers continue to be cited after retraction, sometimes for years, often because citing authors are unaware of the notice. This is why prompt, linked, machine-readable retraction notices matter so much for discovery.

    Can a retracted paper be published again?

    A substantially revised version can sometimes be resubmitted if the authors have genuinely corrected the underlying problem, but this must be done transparently, with the editor informed of the paper’s history. It is never appropriate to resubmit a corrected version without disclosing the prior retraction.

    What this means for editors and research offices

    For journals, precise terminology is a workflow issue as much as an editorial-ethics one: COPE’s flowcharts, ICMJE’s recommendations, and most editorial-management systems expect cases to be tagged with the correct remedy from the outset, because that tag drives downstream indexing signals sent to CrossRef, PubMed, and DOI registries.

    For research offices, the stakes are similar. Case files, funder disclosures, and research-administration compliance reports should mirror the same four-way distinction rather than defaulting to informal language like “the paper was pulled.” Where a case originates in a dispute over who contributed what to a flawed paper, structured contributor statements — the kind increasingly requested under authorship policies — can help institutions establish individual accountability before deciding whether the remedy is a correction or a full retraction. Internal glossaries and training materials can also be cross-referenced against a maintained dictionary of research-administration terminology rather than drafted informally office by office.

    The rise of paper-mill detection tools has also pushed COPE to add explicit guidance on batch retractions — cases where dozens or hundreds of articles from the same source are retracted together for the same systemic reason. That volume makes definitional discipline more urgent: a research office tracking hundreds of cases needs the four categories kept clean to report accurately to funders and institutional leadership.

    Looking ahead

    As journals face more paper-mill-driven batch retractions and more AI-assisted-writing disclosures, the boundary between “correction” and “retraction” will keep being tested in ways COPE’s earlier guidelines did not originally anticipate. Editors and research offices that maintain a precise, shared vocabulary — retraction, correction, expression of concern, and redaction as four distinct tools rather than one blurred category — will be better placed to report consistently, protect the record, and respond quickly when the next systemic case emerges.

  • Retraction Transparency: Why Author Contribution Statements Are Now a Governance Tool

    Retraction in academic publishing has moved from a rare editorial embarrassment to a routine, high-volume governance problem. Springer Nature disclosed that it retracted 1,462 papers across its portfolio in 2025 — a figure that, alongside a January 2026 Nature editorial on the subject, has reframed how publishers, institutions and funders think about the paper trail behind a byline. The editorial’s central argument was not that retraction volumes are alarming in isolation, but that the publishing ecosystem still lacks a reliable, structured way to establish who did what on a paper once something goes wrong.

    That gap is precisely where structured author contribution disclosure now sits. What began in 2014 as a mechanism for fairer academic credit is increasingly being read by editors, integrity officers and institutional research offices as something closer to an audit trail: a documented record of who contributed which specific tasks to a study, retrievable long after publication when a correction, expression of concern or retraction notice becomes necessary.

    This shift matters for research administrators specifically, because it changes what “good practice” looks like at the point of submission. Contribution statements are no longer a courtesy line for the acknowledgements section — they are becoming evidence that institutions and journals may need to produce during a formal investigation.

    Retraction in Academic Publishing: From Rare Event to Routine Governance Signal

    The scale of the Springer Nature figure is instructive. A number in the thousands, drawn from one large multi-journal publisher in a single year, signals that retraction of research papers has become a standing feature of scholarly communication rather than an exceptional event confined to high-profile fraud cases. Retraction Watch has tracked this trend for years through its public database, documenting causes ranging from image manipulation and data fabrication to authorship disputes, undisclosed conflicts of interest and — increasingly — undisclosed or improper use of generative AI in manuscript preparation.

    What the January 2026 Nature editorial added to this picture was a governance argument: publishers cannot investigate at this volume using ad hoc correspondence and memory. Journals need structured, machine-readable metadata about contribution and responsibility captured at submission, not reconstructed after the fact from email threads and co-author recollection. Research integrity issues surface months or years after publication, often when the researchers involved have moved institutions, changed collaborators, or in some cases become uncontactable. A contribution statement recorded at submission time, tied to a persistent identifier, survives all of that.

    Why Contributor Role Taxonomies Function as Audit Trails

    This is where the Contributor Roles Taxonomy becomes relevant to integrity investigations rather than just credit allocation. CASRAI originated the CRediT contributor role taxonomy in 2014. The standard is now stewarded by NISO as ANSI/NISO Z39.104-2022. Its fourteen defined roles — including conceptualisation, data curation, formal analysis, investigation, methodology, and writing (original draft, review and editing) — were designed to solve a credit-allocation problem: junior researchers, data specialists and methodologists were frequently under-recognised by the traditional single-line “authorship” convention.

    The governance value is a by-product of that original design. When a journal applies structured roles at submission, and links each contributor to an ORCID identifier, it creates a queryable record: who claimed responsibility for the statistical analysis, who curated the dataset, who wrote the manuscript. If a subsequent research integrity investigation identifies fabricated data in a specific figure, that record narrows the inquiry to the individuals who claimed the relevant roles — data curation, formal analysis, investigation — rather than treating the full author list as equally implicated or equally responsible. This is precisely the function COPE (Committee on Publication Ethics) guidance has long urged: that authorship and contribution be documented in a way that supports fair, evidence-based adjudication of misconduct allegations, rather than blanket sanctions against every listed author.

    COPE Guidelines and the Documentation Gap

    Retraction guidelines COPE has published over the past decade consistently emphasise two things: that retraction decisions should follow a documented, defensible process, and that all listed authors should be given the opportunity to respond before a notice is issued. Both requirements depend on knowing, precisely, who is responsible for what. In practice, many journals still rely on a single free-text contribution paragraph — “X and Y designed the study; Z performed the experiments” — that is neither standardised nor easily machine-searchable across a portfolio of thousands of papers.

    Structured CRediT statements close that gap. Because the taxonomy uses a fixed, finite set of roles, publishers can query metadata at scale: which papers list a given researcher under “data curation,” across how many journals, in how many retracted studies. Crossref and DataCite metadata schemas already support structured contributor role fields, meaning this information can, in principle, travel with the persistent identifier record rather than remaining locked inside a PDF. That is the technical foundation an audit-trail function requires — and it is largely already in place; the remaining barrier is consistent adoption and consistent metadata deposit by journals and platforms.

    What This Means for Research Administrators

    For institutional research offices, this shift has practical consequences that go beyond publisher policy:

    • Institutional research integrity offices should expect to be asked for contribution records during misconduct investigations initiated by journals or funders, not just the reverse. Retaining structured contribution metadata alongside grant and output records strengthens an institution’s ability to respond quickly and specifically.
    • ORCID linkage is no longer optional infrastructure. With ORCID adoption now effectively mandated across most major funders and publishers, institutions should ensure researcher profiles are current and that contribution claims on outputs are verified, not simply self-reported and forgotten.
    • Authorship disputes should be resolved before submission, using structured role assignment as the basis for discussion, rather than settled informally and revisited only when a correction becomes necessary.
    • Research integrity training should reference contribution statements explicitly, framing them as a professional and accountability record, not an administrative afterthought completed in the final minutes before submission.
    • Institutions preparing for REF 2029 and equivalent national assessment exercises should treat consistent, verifiable contribution metadata as an asset that supports both credit allocation and defensibility should any submitted output later face scrutiny.

    The direction of travel is consistent with wider open science governance trends — UKRI’s evolving open access policy, NIH data sharing enforcement, and Horizon Europe’s research integrity expectations all point toward increased structured disclosure as a condition of funding and publication, not a voluntary enhancement.

    Conclusion: Structured Disclosure as Standard Practice

    Retraction in research has always carried reputational weight for the individuals and institutions involved; what has changed is the expectation that the process leading to a retraction decision be documented, structured and defensible from the outset. The Springer Nature 2025 figure and the Nature editorial that followed it are unlikely to be the last signals in this direction. As integrity investigations grow in frequency and complexity — compounded by emerging challenges around AI-assisted manuscript preparation and image generation — publishers, funders and institutions will continue to look for metadata standards that were originally built for credit and recognition, and increasingly ask them to do double duty as accountability infrastructure. Contribution taxonomies stewarded through formal standards bodies, deposited consistently in persistent-identifier metadata, and linked to verified researcher identifiers are best positioned to meet that demand.

  • Generative AI Authorship: Integrity and Disclosure Standards for Academic Journals

    Introduction to AI Authorship in Scholarly Spaces

    The rapid emergence of Large Language Models (LLMs) and Generative AI tools (such as ChatGPT, Claude, and Gemini) has disrupted academic writing and scientific publishing, raising complex ethical questions regarding authorship, copyright, and research integrity.

    Why Generative AI Cannot Be Listed as a Co-Author

    Major global editorial associations, including COPE, ICMJE, and WAME, have established a strict rule: Generative AI tools cannot be listed as co-authors. Authorship carries legal and ethical responsibilities, including accountability for the accuracy, validity, and integrity of the entire study—responsibilities that a software tool cannot assume.

    Drafting Clear Disclosure Guidelines for Researchers

    While AI cannot be an author, researchers may use AI tools to assist with grammar editing, translation, or literature mapping. Journals must mandate complete disclosure. Authors should state: 1. Which AI tool was used. 2. The specific version and developer. 3. The precise prompt parameters and sections of text or code generated.

    Detecting Unethical AI Writing and Protecting peer review

    Editorial teams are deploying automated detectors to identify undisclosed AI generation. Peer reviewers must also play a role, ensuring that AI-written text does not introduce hallucinations, fabricated references, or biased analysis, safeguarding scientific credibility.

    Key Data and Comparative Metrics

    AI Tool Application Authorship/Disclosure Requirement Standard Manuscript Citation Standard
    Drafting Text / Coding Strictly prohibited as author; must disclose in methodology or acknowledgements. Describe tool use in detail: ‘ChatGPT (v4.o) was used to assist with coding…`
    Grammar / Proofreading Generally allowed without disclosure, as long as original meaning is unchanged. Standard acknowledgement of digital editing tools (optional).
    Image Generation Prohibited in clinical/scientific figures; allowed for conceptual art with clear labels. Cite source tool and license parameters in figure caption.

    Actionable Checklist for AI Authorship

    • Incorporate a strict Generative AI Authorship policy into journal instructions.: Incorporate a strict Generative AI Authorship policy into journal instructions.
    • Mandate a dedicated ‘AI Use Disclosure’ statement in all submitted manuscripts.: Mandate a dedicated ‘AI Use Disclosure’ statement in all submitted manuscripts.
    • Explicitly prohibit listing AI tools as co-authors on submission forms.: Explicitly prohibit listing AI tools as co-authors on submission forms.
    • Train peer reviewers to identify AI-hallucinated citations and arguments.: Train peer reviewers to identify AI-hallucinated citations and arguments.
    • Integrate automated AI writing and translation detection tools into workflows.: Integrate automated AI writing and translation detection tools into workflows.

  • Detecting and Preventing Paper Mills: Ethical Guidelines for Scholarly Publishers

    Introduction to Paper Mills in Scholarly Spaces

    The rise of paper mills—commercial organizations that manufacture fake, plagiarized, or low-quality research papers and sell them to authors—poses a critical threat to the integrity of the scholarly record. Publishers and editorial teams are facing a wave of sophisticated, automated submissions designed to bypass standard peer review.

    Characteristics of Paper Mill Submissions

    Paper mill submissions often share distinct characteristics, including templated titles, recycled datasets, manipulated or stock images, and unexplainable citation spikes. Editorial offices must train staff to recognize these red flags before peer review begins, utilizing automated detection tools to flag suspicious submissions.

    Implementing COPE Guidelines and Best Practices

    The Committee on Publication Ethics (COPE) has developed specific guidelines for handling suspected paper mill operations. Publishers must establish robust triage workflows, cooperate across journals to share fraud signals, and implement multi-step author identity verification protocols during manuscript submission.

    Technical and Structural Anti-Fraud Solutions

    Beyond manual checks, publishers should integrate technical barriers. These include image duplication software, automated text analysis to detect spin tools or generative AI authorship without disclosure, and cross-checking author email domains and institutional ORCID affiliation logs.

    Key Data and Comparative Metrics

    Anomaly Type Paper Mill Indicator Detection Mechanism
    Authorship Anomaly Unexplained, rapid changes in author list or non-institutional emails. Mandatory ORCID verification and historical submission logging.
    Image Manipulation Identical western blots or microscopy grids across different papers. Automated image forensic software scanning (e.g., Proofig).
    Citation Spam Citing unrelated papers from a narrow set of co-authors. Automated citation network analysis and cross-journal reference checks.

    Actionable Checklist for Paper Mills

    • Mandate institution-issued email addresses and validated ORCID iDs for all authors.: Mandate institution-issued email addresses and validated ORCID iDs for all authors.
    • Integrate automated image forensic tools directly into the manuscript submission system.: Integrate automated image forensic tools directly into the manuscript submission system.
    • Establish cross-journal editorial alliances to share data on retracted and suspect authors.: Establish cross-journal editorial alliances to share data on retracted and suspect authors.
    • Train peer reviewers to evaluate raw data and verify the feasibility of experiments.: Train peer reviewers to evaluate raw data and verify the feasibility of experiments.
    • Publish raw datasets alongside peer-reviewed articles to ensure complete transparency.: Publish raw datasets alongside peer-reviewed articles to ensure complete transparency.