# Scientists Defy AI Bans in Peer Review, Raising Questions About Research Integrity
Researchers continue to use artificial intelligence tools for peer review despite explicit instructions to avoid them, according to findings presented to a major conference. The behavior suggests that widespread adoption of AI in science has outpaced institutional controls and researcher compliance with ethical guidelines.
The observation emerged when organizers of an unnamed conference requested that scientists reviewing submitted papers refrain from using AI tools during evaluation. Despite this clear directive, a substantial portion of reviewers ignored the request and deployed AI systems anyway. The researchers declined to specify which conference hosted the experiment or provide exact numbers of violators, citing confidentiality agreements.
This pattern reflects a broader tension in academic science. AI tools like ChatGPT and Claude have become integrated into scientific workflows for drafting, editing, summarizing literature, and generating ideas. Their convenience and utility make them difficult to abandon, even when institutional policies or research ethics demand it. The disconnect between policy and practice reveals how rapidly AI adoption has outpaced governance mechanisms in academia.
The implications for peer review are substantial. Peer review forms the backbone of scientific quality control. When reviewers use AI to process papers, several risks emerge. AI systems can hallucinate citations, miss nuanced methodological flaws, or introduce biases present in their training data. A reviewer using an AI tool to generate their entire critique may produce generic feedback that fails to catch errors or provide constructive guidance that human expertise delivers.
Conference organizers face practical enforcement challenges. Digital monitoring for AI use requires invasive surveillance of reviewer behavior. Watermarking papers or implementing keystroke analysis could detect AI involvement but raises privacy concerns. Many conferences lack technical infrastructure to police compliance. The honor system, as demonstrated by this experiment, proves insufficient when the tools are readily available and their use goes undetected.
The incident also reflects researcher attitudes toward AI governance. Many scientists view AI as a utility rather than a tool requiring ethical gatekeeping. They may rationalize that AI enhances rather than compromises review quality, or argue that the policy is unrealistic given AI's integration into their everyday work. A researcher might spend five minutes using AI to draft a review outline and genuinely not consider this a violation of the spirit of the request.
Several journals and conferences have begun implementing AI disclosure requirements rather than outright bans. Nature, Science, and Cell now require authors to disclose AI use in manuscript generation. Some conferences ask reviewers to acknowledge whether they used AI. These transparency approaches accept that AI use will occur and attempt to make it visible rather than hidden.
The deeper issue concerns research culture and self-regulation. Academic science has long relied on researcher integrity and community norms to maintain standards. When tools become ubiquitous and easy to use covertly, reliance on compliance falters. Stronger measures may include AI literacy training for reviewers, clear consequences for policy violations, and redesigned peer review systems that account for AI's presence rather than pretend it does not exist.
The findings underscore that policy frameworks lag behind technological reality in academic research. Until institutions develop enforcement mechanisms with sufficient credibility and journals establish transparent AI protocols, researchers will continue deploying these tools regardless of formal restrictions.