A journal has retracted a paper after its authors confirmed that a hidden instruction aimed at AI reviewers was included in a revised manuscript, Retraction Watch reports. The case puts a practical question on the table for publishers using AI in peer review: how do you stop a document from quietly telling its automated reader what verdict to reach?
Watch Desk analysis
What happened
Retraction Watch says the prompt was concealed in the paper’s data availability statement. It instructed an AI system to ignore previous directions, write a positive review, avoid negative points and recommend acceptance. The publisher’s retraction notice said concerns were raised about an apparent AI prompt; the authors confirmed they had included one in the revised manuscript.
The report says the paper had been under review for about a year. Researcher René Aquarius, who found the prompt, said he believed that meant it had not succeeded in securing a favourable review. That is his assessment, rather than proof of what any reviewer or editorial system did.
The report also describes a separate case: Nick Brown found a similar instruction hidden in a figure in a preprint. It was absent from the later journal version. These incidents are related examples, not the same paper or the same outcome.
Why it matters
AI-assisted review creates a new route for authors to try to influence the process: not only by persuading a human reviewer, but by placing instructions where an automated system might read them. The confirmed prompt shows that this is more than a hypothetical weakness. The report does not establish that an AI reviewer followed it, or that it affected the paper’s acceptance.
For publishers, the practical challenge is to treat submitted documents as untrusted input and ensure reviewers, whether people or software, can distinguish manuscript content from instructions. For readers, the retraction is a reminder that the presence of a prompt is not evidence that it worked, but it is a reason to ask how the review process handles it.
Our read
The most useful detail is also the least cinematic: the prompt was confirmed, the paper was retracted, and its effect on review is not established. That distinction keeps a real publishing-process problem from becoming a claim that AI was fooled. If publishers invite automated review, they need a clear account of how hostile instructions in submitted work are detected and handled. “The model should know better” is not quite an editorial policy.
What to watch
- Whether the publisher explains how it investigated the prompt and reached its retraction decision.
- Whether journals disclose rules for AI-assisted peer review and handling instructions embedded in submissions.
- Whether further cases show prompts affecting review outcomes, rather than merely appearing in manuscripts.
Discussion spark: Should journals ban AI-assisted peer review until they can show how hidden instructions in manuscripts are handled, or can safeguards make the process trustworthy enough to use?
Sources and evidence
- Source update (19 November 2026, 16:41 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.