NIH bans AI-written grants, awards paused for human re-review

Federal Grant Writing Policies: The National Institutes of Health established that grant applications substantially generated by artificial intelligence do not qualify as an applicant's original work, prohibiting proposals where core drafting is delegated to automated tools. Concurrently, the National Science Foundation allows the use of artificial intelligence in proposal preparation, provided that applicants formally disclose the tool's involvement and assume direct accountability for the factual accuracy of the text.
Award Re-Reviews and Integrity Directives: A research network suspended previously distributed awards to conduct a manual re-review of applications following procedural concerns regarding algorithmic screening. At the institutional level, the University of Utah Office of the Research Integrity Officer issued formal guidance directing investigators to verify all tool-generated statements and citations against primary sources and to adopt laboratory-specific rules for tool usage.
Clinical Trial Operations: Federal agencies launched programs via ARPA-H and the Department of Health and Human Services to adjust clinical trial design, site activation, consent processes, and patient data aggregation. Among these programs, ARPA-H released solicitation ARPA-H-SOL-26-164 for the SURPASS initiative, which seeks consortia of sponsors, sites, and trial arms to test agentic operations and continuous inference models aimed at shortening development timelines.
Ethics and Governance Oversight: The World Health Organization issued recommendations urging expanded resources and technical support for research ethics committees monitoring artificial intelligence in health studies, specifically focusing on algorithmic bias and human rights protections. In parallel, authors released a reference volume addressing governance models, including the NIST AI Risk Management Framework, for managing research administrative workflows and regulatory compliance.
Peer Review and Editorial Procedures: Viewpoint authors introduced the integrated transparency protocol to require disclosure of automated text processing during editorial triage and copyediting, aiming to curb shifts in scientific meaning. A separate narrative review evaluated the integration of automated tools for manuscript screening against potential risks to editorial judgment, while publishing executives maintained that final evaluation still requires human verification.
Regulatory Administration: AstraZeneca restructured its European regulatory affairs operational framework, deploying automated platforms to manage lifecycle tasks, clinical documentation, and submission pipelines between local affiliates and global divisions.

Sources

AI Grant Writing (2026): Funder Rules, Workflow & a Prompt
Use AI to organize, critique and tighten your grant, never to supply its facts or its program design. As of September 2026, NIH won't treat applications "substantially developed by AI" as the applicant's original work, and NSF encourages you to disclose how AI was used and holds you responsible for accuracy. Most private foundations publish nothing, so check each RFP and ask. The workflow and the prompt are below.
How funders treat AI in grant applications - Quillify
Grant funders have varying policies regarding artificial intelligence, with the National Institutes of Health prohibiting applications substantially developed by AI and the National Science Foundation permitting AI assistance provided applicants disclose its use and maintain accuracy. Because most private foundations and other federal agencies have not established agency-wide rules, applicants must consult individual solicitations and guidelines before drafting. When using AI, applicants should keep their program designs and ideas original, rigorously verify all claims and citations, and…
Funder 'sorry' for lack of transparency over AI screening
Research network will undertake human re-review of applications, with previous awards “suspended”
ВОЗ призвала усилить этический надзор за исследованиями в сфере здравоохранения с применением ИИ
Всемирная организация здравоохранения призвала усилить этическую экспертизу и надзор за исследованиями в области здравоохранения с применением искусственного интеллекта для снижения рисков алгоритмической предвзятости и защиты прав человека. Новый доклад организации охватывает весь исследовательский цикл и подчеркивает необходимость расширения ресурсов комитетов по этике, а также участия регуляторов, издателей и грантодателей. Отдельное внимание в документе уделено поддержке исследователей из стран с низким и средним уровнем дохода для предотвращения неравенства и дисбаланса сил в сфере…
US government launches AI-driven programs to overhaul clinical trials
Amid rising industry concerns about U.S. competitiveness with China, the Trump administration is rolling out four ARPA-H initiatives to modernize clinical trial design, site activation, consent and patient data collection.
Generative AI and the Research Record
While generative AI can assist researchers with literature reviews, writing, and organizing ideas, research administrators warn that unverified outputs can introduce fabricated citations, inaccurate claims, or uncredited content. Because researchers remain fully accountable for their submissions under federal research misconduct standards and differing sponsor requirements, tools cannot replace human verification. The University of Utah's Office of the Research Integrity Officer advises researchers to check all claims against original evidence and establish clear, lab-specific expectations…
AI for Biomedical Research Volume 1
Artificial intelligence (AI) is transforming every aspect of biomedical research practice. Researchers face a fragmented landscape of tools, evolving ethical requirements, emerging regulations, and rapidly shifting best practices. The first of two volumes, this book provides a comprehensive treatment of AI across the complete research lifecycle, from initial ideation through analysis, quality assurance, dissemination, and long-term impacts. Key Features This book is the only resource covering AI applications across the entire research journey, from initial idea through long-term impact. It…
HHS launches initiative to bolster US clinical trials market
The federal initiative combines predictive models, shared infrastructure and AI to speed up trial design and reduce operational burdens for researchers.
ARPA-H's SURPASS Wants to Cut Drug Trials From a Decade to Four Years. The Solicitation Asks for Something Almost No Single Organization Has.
ARPA-H's SURPASS program (ARPA-H-SOL-26-164) seeks to shorten clinical drug development to under four years across three required technical areas: a phaseless simulation engine, continuous inference, and an agentic operations layer. The 60-month initiative requires consortia that include at least two for-profit drug sponsors, three intervention arms, and a minimum of six clinical sites, with continuation dependent on securing FDA authorization within a 24-month initial stage. Applicants must submit a mandatory four-page Solution Summary by November 30, 2026, to qualify for invitation to the…
AI Can Screen Research. Can It Be Trusted to Judge It?
In an interview with Scott Douglas Jacobsen, Springer Nature North America President Anna Troise discusses how artificial intelligence, open access, and evolving impact metrics are transforming scholarly publishing. She argues that while AI can streamline manuscript screening and anomaly detection, human editorial judgment and accountability remain essential for evaluating scientific rigor and maintaining trust in the scholarly record. Troise also emphasizes that research assessment should focus on real-world outcomes rather than citation metrics alone, and that publishers must safeguard…
From Compliance to Strategic Partner: The Transformation of Regulatory Affairs in AstraZeneca Local Affiliates
The role of Regulatory Affairs (RA) within local pharmaceutical affiliates is evolving from a predominantly compliance-driven support function to a strategic partner in drug development and patient access. Traditionally, local RA focused on dossier preparation, regulatory compliance, liaison with national authorities, and maintenance of marketing authorisations, largely executing global plans downstream of clinical development. The increasing complexity of novel therapies (including advanced therapy medicinal products), evolving EU regulatory frameworks, and the broader Medical Affairs (MA)…
ARTIFICIAL INTELLIGENCE-ASSISTED PEER REVIEW: A NARRATIVE REVIEW OF CURRENT EVIDENCE, APPLICATIONS, CHALLENGES, AND FUTURE PERSPECTIVES
Background: Peer review remains the cornerstone of scholarly publishing, yet it faces persistent challenges including reviewer fatigue, delays, variability in quality, and increasing submission volumes. Recent advances in artificial intelligence (AI), particularly large language models and automated analytic tools, have prompted growing interest in their potential role in supporting and augmenting the peer review process. Objective: This narrative review critically examines the current evidence on the effectiveness of AI-assisted peer review, highlighting its benefits, risks, limitations, and…
The invisible editor: operationalizing ‘meaningful human oversight’ in artificial intelligence–mediated workflows
Scholarly publishing is undergoing a signiﬁcant structural transition: progressive del-egation of editorial judgement to algorithmic systems. While the debate has focused on authors’ use of generative artiﬁcial intelligence (AI), the undisclosed use of AI by editorial intermediaries – speciﬁcally in the copyediting and peer review triage stages – has received less attention. This Viewpoint presents that opaque editorial AI use risks ‘semantic drift,’ a condition in which algorithmic processing compromises the epistemic integrity of the scientiﬁc record. In response, the integrated…

Also this week

Full transcript

Machine-generated text is no longer recognized as original applicant work in major research proposals. That shift in federal authorship rules leads our coverage today on AI in RA, your monitor of artificial intelligence across research administration. We begin with the policy updates. The National Institutes of Health has put down a direct line on grant applications: substantive development by artificial intelligence is prohibited. And the mechanism there is authorship classification. The NIH rule says text produced by automated models cannot be considered the applicant's original work. If generative software handles drafting, the proposal is disqualified. Which creates an immediate divergence across federal funding, because the National Science Foundation is handling this differently. NSF allows generative assistance in drafting, provided investigators disclose the software and assume responsibility for the text. Right, NSF shifts the burden to validation rather than drawing a boundary at generation. The applicant answers directly for citation integrity, factual errors, and program validity. That validation gap is already showing up at the institutional level. The University of Utah's Office of the Research Integrity Officer just instructed researchers to trace every AI-generated claim back to primary sources, pointing directly to fabricated citations and federal misconduct rules. And the consequences of skipping that manual verification are already visible. One research network had to suspend active grant awards across its portfolio after unverified evaluation tools were used, forcing human re-reviews of all submissions before releasing funds. That tension between automation and manual verification extends directly into clinical development. ARPA-H just issued solicitation ARPA-H-SOL-26-164 for its SURPASS program, aiming to compress drug trial schedules from ten years down to under four. The technical structure is specific: competing consortia have to integrate a phaseless simulation engine, continuous inference, and an agentic operations layer across at least two pharmaceutical sponsors, three intervention arms, and six trial sites. And that runs alongside four broader ARPA-H trial initiatives and a Department of Health and Human Services program targeting site activation, participant consent, and predictive trial design. All driven by competitiveness targets relative to China. On the commercial side, the response looks similar. AstraZeneca restructured its European regulatory affairs operations across three areas, building automated pipelines specifically for lifecycle management, local filings, and clinical evidence. Which raises regulatory compliance questions under frameworks like the EU AI Act and the NIST AI Risk Management Framework. A new two-part volume, AI for Biomedical Research, maps out those operational obligations across study design and ethics. The World Health Organization also flagged this in a new report on health research ethics, pointing out that local institutional review boards, particularly in lower- and middle-income nations, lack the technical resources to audit algorithmic bias. That oversight problem is appearing inside academic publishing as well. In an interview, Anna Troise from Springer Nature pointed out that while automation can handle manuscript intake and anomaly detection, relying on tools for scientific verification threatens trust in the literature. A narrative review examining literature from 2019 to 2026 documented the trade-offs there: automated screening catches plagiarism, but it compromises reviewer judgment, leaks confidential draft data, and carries algorithmic bias. It also introduces semantic drift. A new viewpoint paper outlines protocols to stop automated copyediting and triage tools from subtly altering the underlying meaning of technical manuscripts before peer review even happens. Beyond research infrastructure, OpenAI rolled out visual display ads alongside image generation in ChatGPT for US users, while deploying a virtual clothing try-on feature internationally. In applied analysis, a cybersecurity researcher decoded a Napoleonic military cipher using OpenAI's GPT-6 Astra, and James Wiser reported operational updates from the Kansas State University Libraries AI Stewardship Committee. And new reviews mapped out adoption across distinct sectors: business administration, environmental digital twins for monitoring, border management and customs compliance, and clinical trial pipelines, alongside an organizational study showing AI-generated developmental feedback shifts employee job crafting through affective trust. We will track how these funding requirements evolve next week. From AI in RA, thanks for listening.

From our community

Builders Group

A peer working group for research administrators who are actively building AI into their offices. Some members work alongside IT teams; others are piecing it together on their own. What unites the group is the work of implementation itself:

RSVP →

From our community

Join AI in Research Administration

Join the community →