AI-Assisted peer review: a scoping review of governance, ethical-behavioral risks, and integrity

Generative artificial intelligence (GAI) is reshaping peer review by reducing the cost of producing plausible, well-structured evaluative text while leaving the underlying cognitive work of reviewing only weakly observable. This article argues that AI-assisted peer review is, in practice, difficult to reverse because it is driven by convergent incentives—workload, speed, and competition—in a context where both reviewer effort and AI use are difficult to detect. Drawing on a qualitative analysis of recent literature and institutional policies, the article identifies three interrelated risks: effort outsourcing and accountability laundering, the limited detectability of AI-generated review reports, and adversarial vulnerabilities such as prompt injection embedded in manuscripts. In response, it proposes an auditable hybrid governance model centered on confidentiality, reviewer accountability, secure infrastructure, granular disclosure, and evidence-anchored auditing. The article concludes that preserving peer review as a practice of responsible judgment requires governance mechanisms that make AI-assisted workflows more verifiable without relying solely on bans or textual policing.