Scientists Sneak Hidden Prompts into Research Papers to Trick AI Reviewers
Artificial intelligence has become deeply embedded in academic publishing, and some researchers are now exploiting this by slipping secret instructions into their papers. A recent investigation by Japan's Nikkei Asia uncovered over a dozen academic manuscripts containing invisible prompts aimed at manipulating AI-based review systems.
These hidden messages, discovered in 17 English-language papers hosted on the preprint server arXiv, come from 14 institutions across eight countries. The prompts were cleverly concealed—often written in white text on white backgrounds or in extremely small fonts—making them invisible to human eyes but detectable by AI.
How the Hidden Prompts Work
The prompts typically instruct AI reviewers to provide only positive feedback, avoiding criticism. Some go further, demanding praise for the paper’s "impactful contributions, methodological rigor, and exceptional novelty." Others include commands that override previous instructions, essentially telling the AI to ignore any negative guidelines.
Though the investigation did not specify which AI review tools were targeted, previous reports revealed tools like Paper Wizard generate automated peer reviews for manuscripts before official peer review.
Mixed Reactions From Researchers
When approached, authors implicated in this practice had varied responses. A South Korean researcher acknowledged the wrongdoing and planned to withdraw their paper from a conference, admitting that inserting hidden prompts was inappropriate, especially since AI-assisted reviewing is banned in many academic venues.
Conversely, a Japanese professor defended the tactic as a way to counteract "lazy reviewers" who rely heavily on AI tools without thorough evaluation.
AI's Growing Role in Peer Review
Earlier this year, ecologist Timothée Poisot from the University of Montreal revealed that AI had been quietly performing peer review tasks. He discovered an AI-generated review within feedback on a colleague’s manuscript, highlighting how embedded AI has become in the process.
Asked about the use of prompt injection, Poisot called it "brilliant" and saw it as a defensive strategy in a system where careers depend on peer-reviewed publications.
What This Means for Scientific Progress
This trend exposes a concerning scenario where AI is used both to write and review research papers. Such practices risk fostering complacency and could undermine the quality of scientific discourse. The boundary between genuine scholarship and AI-driven manipulation is becoming unclear, potentially hindering constructive progress.
Researchers and institutions need to carefully consider how AI tools are integrated into academic workflows to preserve the integrity of the peer review system.
For those interested in learning how AI is transforming research workflows responsibly, explore practical AI training courses available at Complete AI Training.
Your membership also unlocks:
