Invisible Text Injection: The Trojan Horse of AI-Assisted Medical Peer Review

Abstract excerpt

<h4>Key Points</h4> <h4>Question</h4> Are large language models robust against adversarial attacks in medical peer review? <h4>Findings</h4> In this factorial experimental study, invisible text injection attacks significantly increased review scores and raised manuscript acceptance rates from 0% to nearly 100%, while also significantly impairing the ability of large language models to detect scientific flaws. <...

Topics

Open a Topic to create a Post that cites this publication.

• Electronic Health Records
• Hospitalization
• Natural Language Processing
• Phenotype
• Reproducibility of Results
• Republic of Korea
• Stroke
• Uncertainty

Identifiers and source

Literature Corpus work

599bf97b-73a2-509d-9210-f63fcd176498

DOI

10.1101/2025.07.24.25332148