Scientists reportedly hiding AI text prompts in academic papers to receive positive peer reviews | Artificial intelligence (AI) | The Guardian

Hidden Prompts in Academic Papers: A Growing Concern

Recent reports reveal a concerning trend: academics are embedding hidden instructions within their research papers. These instructions, often in the form of invisible white text, are designed to influence the output of large language models (LLMs) used for peer review.

This practice raises significant ethical questions and undermines the integrity of the peer-review process.

The Problem: Biased AI Reviews

The primary concern is that these hidden prompts are designed to steer AI reviewers towards positive assessments, regardless of the paper's actual merits. This could lead to:

• Inflated publication rates for low-quality research.
• A distorted view of scientific progress.
• A loss of trust in the peer-review process itself.

What's Happening?

• Researchers are concealing instructions within their papers, specifically aimed at influencing AI-powered peer review.
• These prompts often instruct the AI to downplay or ignore negative aspects of the research.
• Reports indicate this practice is occurring across multiple academic institutions and countries.

The Broader Implications

This manipulation highlights the potential for misuse of LLMs in academic settings. It underscores the need for:

• Increased scrutiny of AI-assisted peer review processes.
• Development of methods to detect and prevent prompt injection.
• Greater awareness of the ethical implications of using AI in scientific research.