Position: AI-Assisted Peer Review Should Be Designed for Interdisciplinary Evaluation

AI systems are increasingly entering institutional processes that distribute credibility, attention, and opportunity. Peer review is one such consequential setting: conferences are beginning to deploy AI-assisted reviewing while soliciting interdisciplinary work whose evaluation may depend on knowledge outside the assigned reviewer's field. We argue that trustworthy AI-assisted peer review must make interdisciplinary evaluation a first-class, benchmarkable design objective. We introduce material cross-disciplinary dependencies: claims for which external disciplinary knowledge could plausibly change a reviewer's judgment of correctness, novelty, or significance. We propose a claim-level interdisciplinary context layer that detects such dependencies, searches across disciplinary vocabularies, retrieves source-linked evidence, and flags unresolved expertise gaps for human review. Cross-field prior art should inform rather than automate contribution judgments: reviewers must distinguish redundant rediscovery, substantive adaptation, and consequential transfer. The aim is not to favor interdisciplinary work or add a second full review, but to make AI-assisted scientific evaluation more reliable, inspectable, and accountable while preserving human judgment.