The change fits in one sentence. An affiliation, "School of Artificial Intelligence, Jilin International Studies University, Jilin, China," had been recorded as a present address for one author, Mingzhe Jin, instead of as a main affiliation. Frontiers in Artificial Intelligence published the fix on July 15, 2026. The document carrying it is a correction notice of roughly sixty-four kilobytes, with two sentences of substance and a line confirming that the original article has been updated. That is the whole of it.
The article being corrected is "Detecting 'large language models fingerprint' for Japanese texts generated by six LLMs." The correction does not touch it beyond the author block. Nothing is retracted. No number is revised, no figure replaced, no conclusion softened or strengthened. Anyone who wants to know what the study found has to go to the original, because the notice reports none of it.
What the record does and does not tell us
Read literally, this document supports very few claims about the science. The title says the researchers looked at Japanese text produced by six large language models, and at something they call a fingerprint: the idea that a writer, human or machine, leaves consistent traces in how they put sentences together. Beyond the title, the only clue is the keyword list printed at the top. It names stylometric analysis, which measures style by counting features rather than reading for meaning; function words, the grammatical connective tissue that carries little content and is hard to control deliberately; parts of speech; phrase patterns; UMAP, a technique for compressing many measurements into a two-dimensional map a person can actually look at; Random Forest, a classifier built from many decision trees voting together; and SHAP, a method for asking which inputs pushed a model toward its answer.
Those are labels, not results. The correction includes no methods section, no sample sizes, no accuracy figures. It would be easy, and wrong, to guess from that keyword list how the pieces were assembled or how well the fingerprinting worked. On all of that, the notice is silent.
What it does record is who did the work and where. Wataru Zaitsu is at the Faculty of Psychology at Mejiro University in Tokyo. Mingzhe Jin, the corresponding author, is now listed with both the School of Artificial Intelligence at Jilin International Studies University and the Institute of Interdisciplinary Research at Kyoto University of Advanced Science. Shunichi Ishihara works in the Speech and Language Laboratory at the Australian National University in Canberra. Satoru Tsuge is at the School of Informatics at Daido University in Aichi, and Mitsuyuki Inaba at the College of Policy Science at Ritsumeikan University in Kyoto. Psychology, informatics, speech science, policy: five authors across three countries, spread over departments that do not usually share a masthead.
The paperwork moved briskly. Frontiers received the correction on June 23, 2026, logged a revision the same day, accepted it on June 29, and published on July 15. The Frontiers Editorial Office approved it. Like the original, it is open access under a Creative Commons Attribution license, which is why its full text can be quoted here at all.
Why it matters
An affiliation line looks like decoration until you consider what it does. It tells readers where to find an author, tells indexing systems which institution to credit, and ties a piece of work to a place that can answer questions about it. A present address and a main affiliation are not the same claim. The first says where someone happens to be now; the second says which institution the work belongs to. Moving one to the other is a small edit with real bookkeeping consequences, and the journal treated it as worth its own publication, its own DOI, and its own permanent identifier in PubMed Central.
That is what most of the correction literature looks like. Retractions get attention; this quieter category, the misplaced affiliation, the transposed initial, the missing funder, is the ordinary maintenance that keeps the published record matching reality. It rarely makes news, and that is roughly the point.
There is one practical warning here for readers. A correction notice is a separate publication from the study it corrects, and it can surface in a database search looking very much like the real thing. This one carries its own DOI, its own PMID, and a title that begins with the word "Correction," but it contains no findings whatsoever. The question underneath it, whether Japanese text written by machines carries signatures consistent enough to identify which model produced it, is genuinely interesting and worth following up. This document simply is not where the answer lives.