AI Review Is a Systemic Risk to Peer Review: Toward a Blockchain-Supported Claim-Level Ledger for Accountability

Peer review is under pressure, and AI review is entering real reviewing workflows. Large language models summarize manuscripts, draft criticisms, check literature, shape rebuttal summaries, and support meta-review, while venues deploy AI feedback and unofficial reviewer-side use is visible. We argue that this shift creates four systemic risks. First, AI review can imitate expert judgment while fabricating evidence, misjudging novelty and significance, or producing fluent but shallow criticism. Second, manuscripts, rebuttals, retrieval contexts, and review workflows can become attack surfaces for hidden prompts, invisible text, contextual poisoning, and multi-stage manipulation. Third, authors may learn to write for AI reviewers, laundering papers into model-preferred styles and strengthening scientific monoculture. Fourth, AI review can dilute accountability by scaling bias, weakening human reviewing skills, and obscuring who formed or verified a judgment. Existing responses, including detection, disclosure, bans, and residual human oversight, are valuable but too coarse for these risks. \textbf{This paper argues that AI review poses major risks to scholarly evaluation and should be governed through a contestable, auditable, and accountable point-by-point AI review process.} We propose AICR-Ledger, a blockchain-supported dispute ledger that decomposes AI-shaped reviews into evidence-bound, author-contestable, reviewer-endorsed, AC-adjudicable claims, and records their lifecycle through a permissioned consortium ledger using hashes, timestamps, state transitions, and pseudonymous signatures. The ledger makes AI-shaped judgments traceable, challengeable, and institutionally accountable.