Reviewers quietly use AI to summarize papers, and authors use it to polish. The line between assistance and fabrication is getting thin. Should journals disclose AI use in review, and has anyone seen a case where it clearly corrupted the process?