A preprint posted to SSRN in July 2026 reports that 401 granted United States patents cite scientific papers that were later retracted. The preprint, by Hana Kim of the Korea Advanced Institute of Science and Technology (KAIST), cross-referenced the PatentsView database of US Patent and Trademark Office (USPTO) patent citations against the Retraction Watch Database, and was first reported by Retraction Watch on July 30, 2026. The preprint has not undergone peer review, and its figures should be read with that status in mind.
What the preprint found
Kim’s analysis identified 401 patents, out of more than 13 million granted US patents, that cite at least one paper now listed in the Retraction Watch Database. The preprint breaks the 401 down by when the citation entered the patent record relative to the retraction: 195 patents cited the paper before it was retracted and before the patent was filed; 92 cited it during examination, meaning the citation was already in the prosecution record while the retraction had occurred or was occurring; and 114 cited it after the patent had already been granted, so the retraction postdates the citation’s appearance in an issued patent.
Kim then narrowed to a smaller set of 39 higher-impact patents — those the preprint treats as more consequential, for example by forward-citation count or commercial relevance — for closer manual review. Of those 39, the preprint reports that 19 relied substantially
on the retracted paper’s findings, as opposed to citing it only incidentally, for background context, or as one of many references in a crowded citation list.
As quoted by Retraction Watch, Kim frames the underlying problem as one of infrastructure rather than intent: Science has correction mechanisms, but correction signals often do not travel into patent records,
and she describes the disconnect between retraction databases and patent-examination workflows as a fixable infrastructure problem.
Why a citation is not the same as reliance
The methodological distinction the preprint itself draws — between a patent merely citing a retracted paper and a patent’s claims actually depending on that paper’s now-discredited finding — is the central caveat in this story, and it is worth stating plainly rather than folding into a headline figure. Patents cite prior art for many reasons that have nothing to do with substantive reliance: to distinguish the claimed invention from what came before, to satisfy a duty of candor to the USPTO by disclosing anything arguably relevant, or because a patent examiner added the reference during prosecution rather than the applicant. A citation appearing in a patent’s file wrapper does not by itself establish that the retracted paper’s result underlies any granted claim.
This is also where a second, unrelated observation becomes relevant. Retraction Watch quotes Boston College law professor Janet Freilich noting that there are a lot of inaccuracies that are permissible in a patent that would not be permissible in a paper.
Patents and peer-reviewed papers serve different legal and evidentiary functions, and a patent’s citation practices — including how thoroughly an examiner or applicant vets a cited source’s current validity — are not held to the same standard as a journal’s citation practices. The 401-patent figure is a citation-linkage count, not, on its own, a count of patents built on fabricated or erroneous science. The 19-of-39 finding narrows that gap somewhat by manually assessing reliance rather than mere citation, but it covers only a small, purposively selected subset of the full 401, not a random or representative sample, so it should not be generalized to the whole set without the same caveat.
Why this matters for technology transfer and research integrity offices
The finding sits at an intersection that neither research-integrity offices nor technology licensing offices monitor by default. Retraction workflows are built around journals, publishers, and the Retraction Watch Database as the community’s de facto index; patent prosecution and patent examination run on prior art searches through USPTO systems that were not designed to flag a cited reference’s subsequent retraction. Once a patent grants, nothing in the standard lifecycle re-checks whether a foundational citation has since been withdrawn.
For a university TTO or licensing office, this is a due-diligence gap worth naming explicitly rather than assuming someone else already checks it. A licensing due-diligence questionnaire for an inbound or outbound deal typically covers freedom-to-operate, inventorship, and prior-art conflicts; it does not typically ask whether the patent’s specification or file history relies on a paper that has since been retracted. Patent attorneys conducting patentability assessments on a new invention disclosure face the same gap in reverse: the retraction-checking step that a careful researcher or journal now runs before publication does not have an equivalent in standard patent prosecution practice.
On the research-integrity side, this preprint extends a running theme: the machinery built to communicate retractions — standards like retraction statements, NISO’s work on communicating retractions and expressions of concern, and community guidance such as COPE’s retraction guidelines — was built for a scholarly-publishing readership. It was not built to interoperate with patent-office citation records, and this preprint is one of the first attempts to measure the size of that gap empirically rather than assert it exists.
What to watch
Three things are worth tracking as this develops. First, whether the preprint is submitted for peer review, and whether the 401 figure, the pre/during/post-grant breakdown, or the 19-of-39 reliance finding change under review — preprint figures on a first pass sometimes shift once reviewers press on methodology, and CASRAI will note here if that happens. Second, whether USPTO, PatentsView, or a retraction-tracking body such as Retraction Watch signal any interest in building a linkage between the two datasets going forward, which is the fixable infrastructure problem
Kim points to. Third, whether any of the 19 higher-impact patents identified as substantially reliant become the subject of a reexamination request, litigation, or licensing dispute that tests whether a retracted underlying finding affects patent validity or enforceability in practice — none has been reported as of this writing.
This is a single, not-yet-peer-reviewed preprint measuring a real and previously unquantified gap between two infrastructures that don’t talk to each other. It is not evidence that patent examination is broadly compromised by retracted science, and it should be read at the scope Kim herself frames it at: a citation-linkage count, a smaller manually-reviewed reliance finding within it, and an infrastructure problem she characterizes as fixable rather than a crisis.
