Invisible Text Injection and Peer Review by AI Models | Synapse
January 18, 2026Open Access
Invisible Text Injection and Peer Review by AI Models
Key Points
The study aims to evaluate how susceptible leading large language models are to invisible text injection in the context of peer review.
Assessed commercial large language models for their vulnerability to text manipulation.
Conducted simulations of the peer review process to test models.
Evaluated the impacts of invisible text on model outputs.
Identified significant vulnerabilities in AI models to invisible text manipulation.
Found that such manipulations can affect the quality of peer review.
Demonstrated that AI's reliance on textual input makes it susceptible to strategic bias.
Abstract
This quality improvement study assesses the vulnerability of leading commercial large language models to invisible text injection manipulation in simulated medical peer review.