The Papers Your Institution Already Published Are the Real Compliance Risk

2026-08-15 · ORBIXER AI LABS

RESEARCH INTEGRITY · ORBIXER AI LABS

The Papers Your Institution Already Published Are the Real Compliance Risk

Screening catches what researchers submit tomorrow. It does nothing for what you already published — and the federal rule that took effect this year doesn't care which came first.

6 min read · Research Deans, VPs for Research, Research Integrity Officers

Most research offices in the United States now believe they have image integrity under control. A screening tool sits somewhere in the submission pipeline. New manuscripts get checked. Editors get flagged. The institution files that under “handled” and moves on.

That belief is built on the wrong denominator.

The screening conversation is almost entirely about new papers — the ones being submitted this semester, this grant cycle, this year. But the exposure that actually shows up in federal proceedings, media investigations, and due-diligence checks on incoming faculty isn't the manuscript sitting in a submission queue. It's the paper your institution published in 2014, indexed in PubMed, cited two hundred times, sitting quietly on PubPeer with an unanswered flag — and never screened at all, because the tools didn't exist yet, or nobody looked.

That backlog doesn't shrink on its own. Most of it never gets corrected. And as of this year, when an allegation surfaces about a paper regardless of its age, the federal government has a formal, binding process it expects your institution to run.

Only 21.5% of PubPeer-flagged papers judged to warrant an editorial notice — a correction, an expression of concern, or a retraction — have actually received one. The rest keep circulating: in the literature, in your researchers' citation counts, in your institution's name.

Part 1: What the evidence actually says

Start with the baseline everyone cites, usually wrong.

In 2016, Elisabeth Bik, Arturo Casadevall, and Ferric Fang published the field's foundation study in mBio: a visual screen of every figure in 20,621 papers published across 40 journals between 1995 and 2014. They found 782 papers — 3.8% of the sample — containing inappropriately duplicated images, and judged that at least half of those showed features consistent with deliberate manipulation, not accidental file mix-ups.

Correcting the record: that study is frequently paraphrased as “1 in 25 papers is fraudulent.” It isn't. The 3.8% figure is the rate of problematic images, not confirmed fraud — and the “at least half” estimate for intentional manipulation applies to that smaller subset, not to the literature at large. The honest number, stated with its denominator, is: of 20,621 screened papers, roughly 1.9% showed signs consistent with deliberate image manipulation. That is still a serious number. It is not the number usually quoted.

The follow-up evidence is more useful than the headline number, because it answers the question institutions actually care about: does screening work? In 2018, the same group audited Molecular and Cellular Biology's own back catalog — 960 papers published between 2009 and 2016 — and compared the manuscript years before and after the journal started screening accepted papers in 2013. Before screening: an average duplication rate of 7.08%. After screening: 3.96%. Screening roughly halved the problem, in the same journal, with the same author pool.

That is the evidence base for a pre-submission integrity check. It is also the evidence base for why pre-submission checking alone is not enough — it only touches papers submitted after the tool is turned on. Everything published before that date is untouched, and it is the part of the collection nobody is auditing.

Part 2: Why this matters to you

The mechanism. A research integrity finding isn't triggered by when a paper was published. It's triggered by when an allegation is received. A 2014 paper flagged on PubPeer in 2026 enters the same institutional process as a manuscript flagged at submission this month — the same sequestration duty, the same inquiry timeline, the same reporting obligation if federal funding is anywhere in its history.

The adaptation problem. Detection tools have improved faster than correction processes have. Deep-learning-based screening (tools like Imagetwin and Proofig use convolutional neural networks trained on biomedical figures) can now flag duplication, splicing, and rotation at a scale no human reviewer could match manually. But a tool that finds the problem does not fix it. Correction still runs through a journal editor, an institutional inquiry, and — if funding was federal — a formal report. The bottleneck was never detection. It's what happens after detection, on papers nobody budgeted time to revisit.

The arms race is administrative, not technical. Every institution that adds pre-submission screening solves the forward-looking half of the problem. Almost none of them budget time or staff to work backward through the existing catalog. That asymmetry is exactly what shows up when a journalist, a competing lab, or a federal auditor runs the same screen your institution never did.

Part 3: The exposure nobody is measuring

Scenario: the faculty hire. A search committee extends an offer. Nobody screens the candidate's back catalog for PubPeer flags before the offer letter goes out. Eighteen months later, a flag surfaces on a paper from their previous institution — and now it's your institution's problem to investigate, on a paper you had no role in producing and no early warning about.

Scenario: the grant renewal. A federal program officer runs a routine check ahead of a competitive renewal and finds an unresolved PubPeer thread on a co-author's earlier paper, unrelated to the current award but attached to the same lab. The renewal conversation now includes a research-integrity conversation nobody scheduled.

Scenario: the media cycle. Institutional image-integrity stories rarely start with the institution. They start with an independent screen — often the same open-source tools your office could be running proactively — surfacing a years-old paper before your office knows it exists.

United States: the rule that changed this year

On September 12, 2024, HHS's Office of Research Integrity issued its first substantial revision to 42 CFR Part 93 — the federal research misconduct regulation — since 2005. The final rule's requirements became applicable to allegations received by an institution on or after January 1, 2026. Institutions had to have revised, compliant policies and procedures submitted with their 2025 annual report, due to ORI no later than April 30, 2026.

The practical effect for a research office: the compliance question is no longer “did we screen this manuscript before publication.” It's “do we have a documented, current process for any allegation that lands on our desk, on any paper, regardless of when it was published” — because that is now the standard ORI will hold you to if a decades-old paper surfaces this year.

United Kingdom: research culture is now a scored line item

REF 2029 replaces the old Environment element with Strategy, People and Research Environment (SPRE), weighted at 20% of the overall REF score — assessed through an institution-level statement (60% of the SPRE score) and unit-level statements (40%). Full submission guidance is due in autumn 2026. Research integrity practice sits inside that environment case: an institution that cannot show a working process for identifying and correcting problematic publications is building a weaker SPRE narrative than one that can.

Status to be verified: confirm the finalized SPRE indicators against Research England's autumn 2026 guidance before citing specifics in an institutional submission.

Part 4: Three questions to run on your own institution

Do we know how many of our own published papers already have unresolved PubPeer flags? Most research offices genuinely don't know. Nobody has ever run the search.

If an allegation lands on a fifteen-year-old paper next month, do we have a documented 42 CFR Part 93–compliant process ready, or would we be building one under deadline pressure? The rule doesn't grade on a curve for papers that predate it.

Would a due-diligence screen on an incoming hire's publication history catch what a journalist would find in an afternoon? If the answer is no, the gap isn't technical. It's that nobody assigned the task.

Part 5: What the solution has to do

Whatever tool or process closes this gap, it has to do four things a submission-only screen does not:

Screen backward, not just forward. Cover the existing catalog of already-published, institution-affiliated papers — not only what's submitted from today onward.

Cross-reference post-publication signals. Check incoming and existing manuscripts against PubPeer flags and known retraction databases, not only against each other.

Produce documentation a regulator will accept. Detection without a reportable audit trail doesn't satisfy 42 CFR Part 93 — the process has to generate the record an institutional inquiry actually needs.

Run at institutional scale without adding headcount. Research integrity offices are small. A process that requires a forensic analyst's time per paper won't get used on a back catalog running into the thousands.

Where ORBIXER fits

ORBIXER Verify is live today, built to check journal and publisher credibility before submission — the forward-looking half of this problem. Portfolio-scale image screening across an institution's existing back catalog, and the institutional reporting layer that would document findings against a 42 CFR Part 93–ready audit trail, are in active development, not yet available. We're not going to tell you the backward-looking half is solved when it isn't. It's the harder half, and it's the one this article is actually about.

What readers can do today

Researchers: Before your next submission, run your own prior figures through an open-source duplication check. Fixing your own record is cheaper than someone else finding it first.

Reviewers: If a figure looks familiar from an earlier paper by the same group, say so in review. It's a two-minute check with real downstream value.

Editors: If your journal isn't screening accepted manuscripts before publication, the MCB data says you're looking at roughly double the duplication rate you'd have with a check in place.

Research offices: Ask, this week, whether anyone owns the task of monitoring PubPeer and retraction databases for your institution's existing output. If the honest answer is “nobody,” that's the gap to close first — before the next allegation makes the decision for you.

The author is an IIT Kharagpur alumnus and Founder of ORBIXER AI LABS.

Sources

Bik, E.M., Casadevall, A., & Fang, F.C. (2016). The Prevalence of Inappropriate Image Duplication in Biomedical Research Publications. mBio, 7(3), e00809-16.

Bik, E.M., Fang, F.C., Kullas, A.L., Davis, R.J., & Casadevall, A. (2018). Analysis and Correction of Inappropriate Image Duplication: The Molecular and Cellular Biology Experience. Molecular and Cellular Biology.

Ortega, J.L. (2022). Classification and Analysis of PubPeer Comments: How a Web Journal Club Is Used. Journal of the Association for Information Science and Technology.

U.S. Department of Health and Human Services, Office of Research Integrity. Final Rule, 42 CFR Part 93, Federal Register, September 17, 2024.

Research England / REF 2029. SPRE Guidance and Initial Decisions, December 2025–2026.