AI‑Generated Text in Academic Papers: Cases, Retractions, and Policy Guidance

Recently, a physics paper unintentionally included a ChatGPT interface button, and the manuscript passed peer review before being published, highlighting how AI‑generated text can slip into scholarly articles.

The phenomenon was first spotted by Guillaume Cabanac, a deputy professor at the University of Toulouse, who noted that the paper contained the phrase “Regenerate response” on its third page, a clear sign that the authors used ChatGPT to draft parts of the manuscript without disclosure.

Publishers have since retracted the paper for violating ethical policies that require authors to declare AI assistance. Similar cases have been identified through PubPeer, with dozens of articles containing AI‑generated prompts such as “Regenerate response” or “As an AI language model, I …”.

These incidents pose challenges for reviewers, who often lack the time to detect AI‑generated content, and for the scientific community, which must develop methods to identify fabricated references and other AI‑induced errors.

Despite the risks, many publishers, including Elsevier and Springer Nature, state that using AI tools is permissible provided the authors disclose their use. For example, radiology researcher Som Biswas openly used ChatGPT to write sixteen papers, five of which were published, after declaring AI assistance.

This illustrates that, when transparently disclosed, large language models can boost productivity across disciplines, though the community must balance efficiency with rigorous ethical standards.