The Surge of AI in Scholarly Peer Review

Artificial intelligence tools are transforming how manuscripts are evaluated in academic publishing. In the United States, where leading research universities and federal agencies set many standards for scholarly communication, the rapid adoption of these technologies has sparked urgent discussions about the need for clearer, more consistent policies. A recent survey of 1,600 academics across 111 countries found that more than half have already used AI tools while reviewing manuscripts, often without explicit institutional guidance.

US-based publishers and universities are responding with new guidelines that stress confidentiality, human accountability, and transparency. Major outlets such as Taylor & Francis explicitly state that peer reviewers must not upload unpublished manuscripts or portions of them into generative AI tools, while allowing limited use for language polishing only if the reviewer retains full responsibility for the final report. Similar restrictions appear in policies from Elsevier, Springer Nature, and university presses including the University of North Carolina Press and the University of Texas Press.

Why Policies Are Evolving Now

The pace of change has accelerated since 2025. Data from Springer Nature indicate that in 2025 alone more than 1.5 million papers received support from nearly 60 AI tools focused on screening, editorial assessment, and research integrity checks. Projections suggest a further 25 percent increase in 2026. At the same time, a large-scale analysis of more than 5 million papers revealed that while roughly 70 percent of journals now have some form of AI-use policy, these guidelines have not significantly slowed AI adoption or improved disclosure rates. Only about 0.1 percent of post-2023 papers in one sample explicitly disclosed AI assistance.

Federal agencies have taken firm positions. The National Institutes of Health and the National Science Foundation prohibit reviewers from using generative AI tools to analyze grant applications or manuscripts, citing risks to confidentiality and the integrity of expert judgment. University-level guidance at institutions such as Columbia University and the University of Michigan echoes these restrictions for internal peer-review processes.

Core Principles Emerging in US Guidelines

Across high-impact journals, three recurring principles dominate new AI policies: accountability, confidentiality, and maintenance of review standards. Accountability requires that human reviewers remain ultimately responsible for every assessment, even when AI assists with language or organization. Confidentiality rules universally bar uploading any part of a submitted manuscript to external AI platforms. Standards emphasize that AI cannot replace the nuanced expertise required to evaluate methodology, novelty, and scholarly contribution.

Disciplinary differences are noticeable. Science, technology, and medicine fields tend toward stricter prohibitions, while humanities and social sciences journals sometimes adopt more flexible language around assistive use. A cross-disciplinary study of 802 journals found that 83 percent of high-impact-factor titles now maintain explicit AI guidelines for peer review, up from 77 percent earlier in 2025.

Real-World Examples from US Publishers and Universities

Academic Medicine and MedEdPORTAL introduced detailed reviewer guidance in 2026 that prohibits uploading manuscripts to AI tools while permitting reviewers to use AI for improving the clarity of their own written reports. The Foundation for Food & Agriculture Research issued a standalone policy in late 2025 that bars any use of generative AI in its grant peer-review process to protect proprietary proposal content.

University presses are aligning with these trends. UNC Press updated its policy in April 2026 to state that reviewers generally may not employ AI tools to evaluate projects or draft reports. The University of Texas Press requires authors to disclose AI use at submission and includes similar expectations for reviewers in its journal portfolio.

Photo by Erik Mclean on Unsplash

Challenges in Implementation and Enforcement

Despite widespread policy adoption, enforcement remains difficult. Many reviewers work independently and may not always check the latest journal guidelines. Detection tools for AI-generated text produce high false-positive rates, making verification unreliable. Some researchers report using AI privately for summarization or language improvement without disclosure, creating an uneven playing field.

US higher-education institutions are exploring training programs and internal review checklists to address these gaps. Librarians and research integrity offices at universities such as Purdue and UC Davis have developed guides that help faculty navigate publisher-specific rules before submitting or reviewing manuscripts.

Benefits and Risks of AI-Assisted Review

Proponents argue that AI can accelerate the review process by flagging statistical inconsistencies, checking references, or identifying potential image manipulation. Tools already assist editors with initial screening at several large publishers. When used responsibly, these systems may reduce reviewer fatigue and shorten turnaround times for authors.

Critics highlight risks to intellectual property, potential bias in AI outputs, and the erosion of the human expertise that underpins peer review’s credibility. Confidentiality breaches could expose unpublished ideas to training datasets, while over-reliance on AI might diminish the critical judgment that distinguishes high-quality scholarship.

Stakeholder Perspectives Across US Academia

Faculty members at research universities often welcome assistive tools for language editing but remain wary of any policy that appears to replace expert evaluation. Graduate students and early-career researchers express concern that inconsistent rules could disadvantage those less familiar with emerging technologies. Journal editors stress the need for uniform standards across publishers to reduce confusion. Federal program officers emphasize protecting the confidentiality of grant applications as a non-negotiable priority.

Looking Ahead: Recommendations for Updated Policies

Experts recommend that US institutions and publishers move beyond declarative statements toward verifiable frameworks. Suggested steps include mandatory training for reviewers, standardized disclosure language, and clearer distinctions between permissible assistive uses and prohibited full automation. Collaboration between major publishers, university associations, and federal agencies could produce model policies adaptable across disciplines.

Some advocates propose pilot programs that test AI tools under controlled conditions with human oversight, generating data on effectiveness and risks. Others call for greater investment in open-source review platforms that keep sensitive content within institutional firewalls.

Photo by Evgenii Vasilenko on Unsplash

Implications for US Higher Education Careers

As policies solidify, faculty and researchers will need to demonstrate familiarity with responsible AI practices during hiring, tenure, and promotion reviews. Training in research integrity now increasingly includes modules on AI ethics. Universities that provide clear internal guidelines and support may gain advantages in attracting and retaining talent committed to high standards of scholarly communication.

Actionable Steps for Researchers and Administrators

Researchers should consult the specific AI policy of every journal or funding agency before submitting or agreeing to review. When in doubt, default to full disclosure and avoid uploading any confidential material to external platforms. Administrators can strengthen institutional capacity by hosting workshops, updating research integrity handbooks, and coordinating with library staff on policy navigation resources.

Professional associations and university consortia are well positioned to develop shared training modules and best-practice repositories that benefit the broader US academic community.