From Policy to Architecture: A Review of Confidential, Human‐Accountable AI Support for Scholarly Peer Review

ABSTRACT Generative AI tools are increasingly used by peer reviewers, yet editorial and publisher policies address this almost entirely through behavioural guidance, leaving open how confidentiality, accountability and disclosure obligations affect the reliability of the published scientific record. Whether these obligations can be translated into technical design requirements for reviewer‐support systems has not been synthesised. This structured narrative review searched PubMed/MEDLINE, Scopus, Google Scholar, arXiv and official policy pages (June 2026). Fifty‐nine sources were reviewed in total; 48 were formally cited, comprising 19 scholarly articles, 16 arXiv preprints (treated as emerging technical risk evidence, not confirmatory findings) and 13 policy and governance documents—11 from nine editorial organisations and publishers and two broader governance instruments. Evidence was synthesised across 13 inductively identified themes (protocol: OSF 10.17605/OSF.IO/BH4TS ). Policies from nine organisations address AI in peer review but remain predominantly behavioural; none provided a complete technical specification for auditability or enforcement, and only two referenced partial technical measures. Documented risks include hallucination, confidentiality breach through external upload, automation bias, hidden AI use and prompt injection. A consistent implementation gap exists: obligations on confidentiality, accountability and disclosure are not encoded into reviewer workflows. Twelve architecture‐level design principles—local‐first processing, deterministic orchestration and auditable disclosure trails—translate these obligations into structural workflow properties; a local‐first workbench illustrates their operationalisation. Governable AI‐assisted peer review requires governance by design, not merely trust in reviewer restraint. Publishers, journals and editorial organisations should move beyond behavioural guidance to specify technical expectations for reviewer‐support tools; empirical validation remains a priority.

Authors

Institutions

Publication Details

Journal
Learned Publishing
Published
2026-11-16
DOI
https://doi.org/10.1002/leap.2119
Primary Topic
Ethics and Social Impacts of AI
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

From Policy to Architecture: A Review of Confidential, Human‐Accountable AI Support for Scholarly Peer Review

Sharon Baisil, Sanjay B. Kini, Urvish Joshi, Het Joshi
2 links
Learned Publishing
Ethics and Social Impacts of AI
article

From Policy to Architecture: A Review of Confidential, Human‐Accountable AI Support for Scholarly Peer Review

Sharon Baisil, Sanjay B. Kini, Urvish Joshi, Het Joshi
article en

Abstract

ABSTRACT Generative AI tools are increasingly used by peer reviewers, yet editorial and publisher policies address this almost entirely through behavioural guidance, leaving open how confidentiality, accountability and disclosure obligations affect the reliability of the published scientific record. Whether these obligations can be translated into technical design requirements for reviewer‐support systems has not been synthesised. This structured narrative review searched PubMed/MEDLINE, Scopus, Google Scholar, arXiv and official policy pages (June 2026). Fifty‐nine sources were reviewed in total; 48 were formally cited, comprising 19 scholarly articles, 16 arXiv preprints (treated as emerging technical risk evidence, not confirmatory findings) and 13 policy and governance documents—11 from nine editorial organisations and publishers and two broader governance instruments. Evidence was synthesised across 13 inductively identified themes (protocol: OSF 10.17605/OSF.IO/BH4TS ). Policies from nine organisations address AI in peer review but remain predominantly behavioural; none provided a complete technical specification for auditability or enforcement, and only two referenced partial technical measures. Documented risks include hallucination, confidentiality breach through external upload, automation bias, hidden AI use and prompt injection. A consistent implementation gap exists: obligations on confidentiality, accountability and disclosure are not encoded into reviewer workflows. Twelve architecture‐level design principles—local‐first processing, deterministic orchestration and auditable disclosure trails—translate these obligations into structural workflow properties; a local‐first workbench illustrates their operationalisation. Governable AI‐assisted peer review requires governance by design, not merely trust in reviewer restraint. Publishers, journals and editorial organisations should move beyond behavioural guidance to specify technical expectations for reviewer‐support tools; empirical validation remains a priority.

Learned PublishingVol. 40(1)
Northeastern University (US), Manipal Academy of Higher Education (IN), Malankara Orthodox Syrian Church Medical College Hospital (IN), B.J. Medical College (IN)
Peace, justice, and strong institutions
Ethics and Social Impacts of AI
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon