Abstract
Generative AI can draw explanatory figures and can also fabricate research images. We asked how the 100 highest-impact journals regulate AI-generated images and how those rules changed in a year. On 30 September 2026 we captured each journal’s author instructions and the publisher policies they cite. We retrieved archived versions from September–October 2025 and coded rules for five image types. At baseline, 79 journals belonged to publishers that prohibited AI-generated images. Between May and August 2026, Taylor & Francis, Elsevier and Springer Nature replaced these bans with rules that depend on image type. By September 2026, 55 journals explicitly permitted AI-generated explanatory illustrations and 58 permitted AI-assisted data visualisations, usually with disclosure or reproducibility conditions. For 89 journals, AI-created or AI-altered primary research images were prohibited; no journal permitted them. Rules for graphical abstracts and covers varied most. The Science journals, The Lancet family and JAMA Psychiatry still restricted illustrative uses. Twenty-seven journals gave instructions that contradicted or lagged behind their publisher’s current policy, including 20 of 31 Springer Nature journals. Authors need to check journal-level instructions, and publishers need to align them.