What is it about?
Some researchers have begun hiding instructions inside academic manuscripts, using text that human readers may not notice but AI systems can still read. These hidden prompts tell AI tools to give the paper a positive review or recommend acceptance. This article explains how such prompts work, why they matter for peer review, and why they should be treated as a serious research-integrity problem. As journals and conferences increasingly face reviewer fatigue and growing pressure to use AI tools, hidden prompts reveal a new weakness in scholarly publishing. The article argues that academic institutions, publishers, and journals need clearer rules, better detection tools, and more responsible ways to use AI in peer review without allowing authors or reviewers to manipulate the evaluation process.
Featured Image
Photo by Kurt z on Unsplash
Why is it important?
This article is timely because AI tools are already entering peer review while journals and conferences are still deciding how they should be used. It is unique in documenting a concrete new tactic: authors hiding instructions inside manuscripts to influence AI-generated reviews. Rather than treating this as an isolated trick, the article shows why hidden prompts reveal a broader weakness in scholarly publishing: any AI system that reads academic papers can potentially be manipulated by instructions hidden in those papers. The work may help journals, publishers, and research institutions respond earlier by improving screening tools, clarifying AI-use policies, educating researchers, and designing safer ways to use AI without weakening human accountability. The goal is to protect trust in peer review before these tactics become more common, more sophisticated, and harder to detect.
Perspectives
I wrote this article because the hidden-prompt episode struck me as an early warning signal. As AI tools move into peer review and other parts of scholarly publishing, we are creating systems that can read academic texts but may also be manipulated by them. My concern is about recognizing the broader vulnerability before it becomes routine. Peer review already depends on trust, judgment, and imperfect human labor. AI may help with some of that work, but only if we build clear rules, technical safeguards, and real accountability around its use. I hope this article encourages journals, conferences, publishers, and researchers to treat AI-assisted evaluation as a governance problem, not simply a convenience.
Zhicheng Lin
Yonsei University
Read the Original
This page is a summary of: Hidden Prompts in Manuscripts Exploit AI-Assisted Peer Review, Communications of the ACM, June 2026, ACM (Association for Computing Machinery),
DOI: 10.1145/3779116.
You can read the full text:
Contributors
The following have contributed to this page
