Shigeki Matsubara
Most journals' and publishers' guidelines regarding the use of artificial intelligence (AI) in peer review contain a similar statement: "reviewers are prohibited from uploading the manuscript to software or AI-assisted tools or technologies where confidentiality is not assured." I believe that reviewers should refrain from uploading not only the submitted manuscript but also their own review sheets to AI tools. This position is based on two reasons. First, even if AI tools state that uploaded data will be deleted upon request, actual deletion cannot be independently confirmed. Second, reviewers sometimes include highly confidential keywords or messages in the review sheet; uploading them may jeopardize confidentiality. My humble experiment indicates that even when deletion of uploaded data is requested, AI systems such as ChatGPT may retain materials for training purposes; thus, complete deletion is theoretically impossible. Authors are usually permitted to upload their own manuscripts for linguistic refinement. Although this may also jeopardize confidentiality, authors do so voluntarily and accept that risk. Reviewers, however, are entrusted with unpublished work and should therefore exercise far greater caution when considering the use of AI software. Editors and journals expect reviewers to evaluate the significance of the study and manuscript. The review sheet can, and perhaps should, be simple and straightforward, written in their own English. Because confidentiality cannot be independently verified and complete deletion cannot be guaranteed, reviewers should refrain from uploading manuscripts or review sheets to AI tools.