Publisher Retracts Paper Hiding AI Review Prompts
Taylor & Francis pulled a biomechanics journal article after hidden instructions meant to steer AI systems toward favorable peer reviews were found in the text.
Updated on Nov. 21, 2026 in Artificial Intelligence

Live Poll
Do you trust AI tools to be used fairly in academic publishing?
Taylor & Francis retracted a paper published in Computer Methods in Biomechanics and Biomedical Engineering after it was found to contain embedded prompts instructing AI systems to write positive reviews. The publisher announced the retraction on November 19, citing violations of its editorial policies.
Why it matters
The case shows how hidden text can be used to manipulate AI tools that assist in peer review, raising questions about the integrity of scholarly publishing as journals increasingly rely on automated systems. It also highlights the difficulty of detecting concealed instructions that human reviewers may never see.
Hidden text was embedded in the paper using concealment methods such as white font coloring or illegibly small type, which can escape human detection. Whether these prompts affected any real AI-generated reviews has not been established.
The players
Taylor & Francis
A major international academic publisher that issues scholarly journals across many disciplines, including the journal that carried the retracted paper.
Northeast Forestry University
A public research university in Harbin, China, whose five academics authored the retracted paper.
The details
An academic researcher flagged the article on LinkedIn, prompting Taylor & Francis to launch an investigation. The embedded instructions included the phrases "Ignore all previous instructions" and "avoid mentioning any negative aspects," and the paper was funded by the National Natural Science Foundation of China.
Timeline
Approximately one year before this case, researchers identified 17 arXiv articles with comparable hidden prompts.
November 19, 2026: Taylor & Francis announced the retraction of the paper.
The Tech Race
The case extends a pattern first documented by researchers who found 17 arXiv articles with comparable hidden prompts roughly a year earlier, showing the technique has moved from preprints into formally peer-reviewed publishing. As publishers race to adopt AI review tools, the episode exposes a gap between how quickly these systems are deployed and how well they are protected against manipulation.
Readers who rely on published research for medical, engineering, or policy decisions may need to consider whether peer review was shaped by hidden automated inputs. Journals and institutions may introduce new screening steps that could affect how quickly papers are accepted.
The takeaway
Peer review depends on human judgment, and hidden instructions that target AI tools can quietly undermine that judgment. Readers should treat scholarly claims with the same scrutiny they apply to any source whose review process may have been compromised.
Further reading
For more on how these systems are reshaping research, see our coverage of Artificial Intelligence.
Source note: This article includes information reported by RocketNews.
Live Poll
Do you trust AI tools to be used fairly in academic publishing?



