IntraGuard: A committee-side defense against outsourcing review to commercial chatbots

The paper proposes IntraGuard, a defense framework for editorial boards and program committees to prevent reviewers from outsourcing peer review entirely to commercial chatbots. The method exploits the property that structure and visual presentation in a PDF can be separated, embedding heterogeneous defensive text objects in the underlying PDF structure while not changing the visual presentation of the page; there are three in-flow injection mechanisms for deployment. It supports two types of strategies: an explicit strategy that triggers refusal or warning signals, and an implicit strategy that embeds predefined text markers into the generated review. On a total of 17,844 cases across 7 real commercial chatbot settings and conference or journal scenarios in 12 different disciplines, the defense success rate is up to 84%, while keeping human reviewers' reviews unaffected; the paper also evaluates 11 adaptive attacks targeting manuscript sanitization and instruction interference, and discusses the implications of building integrated defenses.