Artificial Intelligence in Scholarly Peer Review: Applications, Benefits, Risks, and Recommendations
Health Innovation Reports · 2026
DOI: 10.64048/hir.v2n2.004 ↗Abstract
Background: Artificial intelligence (AI) use is growing in scholarly peer review to improve efficiency, but its adoption raises concerns. This narrative review examines current and emerging applications of AI in scholarly peer review, synthesizes reported and anticipated benefits, evaluates scientific, ethical, and organizational risks, and offers recommendations.Methods: A targeted narrative synthesis was undertaken of empirical studies, literature reviews, scholarly commentaries and editorials, research-integrity guidance, publisher guidance, and an AI-governance preprint. Evidence was organized according to the editorial workflow and interpreted using four considerations: consequence of error, data sensitivity, explainability and auditability, and the contextual scientific judgment required.Results: AI applications include administrative triage, manuscript classification, reviewer identification, language and consistency checking, integrity screening, manuscript summarization, review-report support, and editorial synthesis. These may reduce routine workload, accelerate selected processes, improve communication, and broaden reviewer discovery. However, evidence that general-purpose AI can reliably assess validity, novelty, significance, ethics, or publishability remains limited. Major concerns include hallucination, fabricated citations, automation bias, demographic and linguistic bias, opacity, confidentiality and intellectual-property breaches, plagiarism, deskilling, unequal access, epistemic homogenization, and displacement of scholarly judgment.Conclusion: AI should be implemented as a bounded, transparent, and auditable assistant rather than as an autonomous reviewer or decision-maker. Journals should adopt risk-based policies emphasizing confidentiality, transparency, human accountability, validation, fairness, and continuous monitoring. The most defensible model is hybrid peer review in which AI supports information-intensive, reversible tasks while human experts retain responsibility for scientific judgment and editorial decisions.
EUVIMED