Potential AI Fabrication of Neuroscience Studies Raises Concerns Over Open Neuroimaging Datasets

When Dorothy Bishop was asked to review a paper for a small neuroscience journal in August, she recognized from the abstract that the authors had used a dataset she had first published four years earlier on the public OpenNeuro repository. The dataset comprised raw functional MRI scans from fifty participants recorded while they performed language‑related tasks such as word generation and picture‑word association.

“I was curious that somebody had already analyzed it, because it’s a sizable dataset that would require considerable effort to examine,” Bishop, an emeritus professor of developmental neuropsychology at the University of Oxford, explained to Retraction Watch.

During a careful read‑through Bishop identified serious inconsistencies between the description of the dataset and the activities she and her colleagues had actually carried out, and several citations failed to substantiate the claims made in the paper. “There was clearly something seriously wrong,” she said. The editor expressed suspicion, and the manuscript was eventually rejected. Nevertheless, Bishop suspected the authors intended to find another outlet for the work, prompting a wider search for related publications.

The broader investigation uncovered two papers from this year in the Journal of Neurolinguistics, an Elsevier journal. Both claim to employ brain‑imaging data from OpenNeuro, yet Bishop and fellow researcher Kathy Rastle—representative of data appearing in one study—conclude that each appears to have been fabricated by artificial intelligence.
Continue reading below…

Her continued interest led to the discovery of the two questionable articles, which now spark debate about manuscript authenticity.

“I was flabbergasted that someone could spend time to generate a paper that I believe is fake,” Rastle told us intently.

Although the papers are presented persuasively, the authors do not indicate a thorough analytical approach beyond loading a modest portion of the data to produce figures. Fundamental methodological errors emerge, suggesting deep inauthenticity. Bishop has formally requested the journals’ retraction, and Elsevier has announced an ongoing investigation.

The first article, published in May and co‑authored by Afzal Khan, Dawood Khan of King Saud University (Saudi Arabia), and Iqra University (Pakistan), purports to use the publicly documented dataset created by Rastle and collaborators. When examined closely, the researchers concluded that the paper almost certainly never employed their data at all.

The manuscript describes behavioral metrics as available only for some participants when in reality they were recorded for every participant. It also reports reading‑comprehension scores absent from the original dataset—a discrepancy that undermines its validity. Moreover, a diagram supposedly illustrating brain regions involved in reading bears no resemblance to established neuroscientific knowledge, causing immediate skepticism among experts.

The paper’s acknowledgments laud Rastle and her team for sharing their data, yet Rastle stresses that she believes the contributors merely invented the impression of legitimacy rather than genuinely using the original dataset.

The second study, released in August by Afzal Khan alone, alleges analysis of four OpenNeuro collections. Rastle flagged concerns on PubPeer, and the author disputed the accuracy of the claims. The reported participant count did not align with the underlying datasets; Khan later stated he corrected participant exclusions, though the revised numbers still conflicted with the original data, according to Bishop’s posted comments.

A misleading graphic depicted a participant as having used a second language for over fifty years, despite the maximum age value in the source material being thirty‑seven. Khan defended the representation on PubPeer, insisting the plot showed no values above fifty. He ultimately acknowledged the need for revised analyses and indicated that a correction process was underway with the journal.

“This is evidently sophisticated AI capable of producing a plausible paper using familiar data sources,” Bishop remarked. “The superficial coherence can mimic genuine scholarship, yet subtle misstatements reveal the role of AI in fraudulent research.” She advises scholars to verify datasets, code, and references, emphasizing that current validation methods lag behind the quality of modern fabrications.

Khan affirmed that he fully acknowledges the concerns and is engaging the editorial office in good faith. He declined to speak specifically while a formal review was pending, assuring that appropriate scholarly scrutiny will occur.

Following Bishop’s appeal, Elsevier confirmed the manuscript would be examined thoroughly, potentially requiring several months before any resolution. Nai Ding, editor of the journal, noted that the authors had submitted findings of their internal inquiry to Elsevier, which stated it was conducting additional review.

“When a paper is demonstrably deeply flawed—or, as here, apparently fake—it should be removed promptly, lest it contaminate the scientific record and confuse peers,” Rastle observed. “Such cases undermine trust in open‑data practices that rely on widespread reproducibility.”

Khan holds a Ph.D. in linguistics and is affiliated with the English Language Skills Department at King Saud University in Riyadh. Prior to 2026 his academic profile focused on linguistic theory, teaching English, and literary criticism, with no prior contribution to neuroscience imaging research. Bishop points out that advanced brain‑imaging projects demand substantial computational resources and specialized expertise—incredible considerations absent from Khan’s background.

The episode provokes strong reactions among advocates of open data, as the unauthorized reuse of participant‑level information threatens transparency and credibility within the discipline.

Ultimately, the case underscores the paramount importance of rigorous validation, independent verification, and vigilance against misinformation masquerading as legitimate scientific work.

Authored by Lori Youmshajekian. First published at Retraction Watch.