ORI AI misconduct guidance, Rice strategy, GenAI grant drafting

The Office of Research Integrity released guidance instructing research institutions on how to handle research misconduct proceedings involving artificial intelligence, defining protocols for inquiries where automated systems are linked to data fabrication, falsification, or plagiarism.
Rice University outlined an institutional framework across operations, research, and instruction, pairing centralized resources with department-level decision-making to automate administrative workloads while maintaining human oversight.
Research administration offices are using generative tools for internal drafting and procedural checklists, with institutional policies mandating human review to comply with federal sponsor rules that prohibit entering proprietary grant data into external platforms.
Grant applicants face fragmented funder policies regarding automated tools: the National Institutes of Health excludes proposals substantively prepared by artificial intelligence from being treated as original applicant ideas, the National Science Foundation requests disclosure, and a Candid survey found that most philanthropic foundations have not established formal rules.
Compliance guidance for research institutions outlines a risk-tiered model for automated tools, advising offices to confine automation to low-level functions like pre-submission checks while preserving human authority for regulatory committee decisions and vetting third-party data security.
The White House announced $2.4 billion in compute commitments for the Genesis Mission Consortium.
The Department of Homeland Security published a proposed rule to establish fees for Optional Practical Training recommendations.
Uber deployed ChatGPT Work across its workforce.
The federal government shifted official terminology from Artificial Intelligence to Super Intelligence.
The Allen Institute for AI launched the LitPivot ideation tool.
OpenAI expanded its storage architecture to support over one billion users.
AT&T implemented ChatGPT to assist with internal legal knowledge retrieval.

Sources

Reg Roundup 1.1: News You May Have Missed + AI Policy Movement - HRP Consulting Group
This article reviews recent federal research policy developments, including NIH updates to K-award requirements and biosafety oversight, an FDA transparency initiative for inspection reports, upcoming FDA guidance on clinical trial enrollment action plans, and planned HHS rulemaking on graduate medical education accreditation. It also examines emerging federal policy on artificial intelligence in research, detailing FDA's regulatory framework for generative AI-enabled medical devices, a proposed NIH Office of the Chief Artificial Intelligence Officer, and ORI guidance addressing AI in…
AI fireside chat: Key takeaways on future of AI at Rice
Rice University launched AI Week 2026 with an executive fireside chat discussing how artificial intelligence will shape the institution's future across research, teaching and campus operations. University leaders emphasized that AI adoption should remain human-centered, ethical and sustainable, utilizing the technology to automate administrative tasks and enhance learning without compromising academic rigor or personal connections. The panel highlighted priorities including campuswide AI fluency, adaptable classroom policies guided by faculty, and governance that balances universitywide…
AI in Research Administration: Practical Workflows, Sponsor Restrictions, and Human Checkpoints
Where generative AI can assist with internal research administration drafts and checklists — and where federal sponsor confidentiality rules, peer-review prohibitions, and human verification draw a hard line.
AI in Grant Writing: What Funders Allow | Aventis West
AI helps with the reading, sorting, and editing parts of grant writing. It hurts when it writes your needs statement, invents data, or flattens your voice. NIH will not treat applications substantially developed by AI as the applicant's original ideas, NSF encourages disclosure, and most foundations in a 2024 Candid survey were still undecided.
AI Readiness for Research Compliance: Where to Start
Research institutions preparing to integrate artificial intelligence into research compliance must first establish clear governance, reliable data infrastructure, and human accountability to ensure decisions remain defensible under federal oversight. Implementation priorities should rank use cases by regulatory risk, starting with low-impact administrative tasks like pre-submission checks while keeping formal committee determinations with qualified personnel. Successfully navigating adoption also requires evaluating technology vendors on data privacy and security, as well as testing tools…

Also this week

Full transcript

When research fraud involves algorithmic tools, federal oversight faces an entirely new set of evidentiary standards. That's what we examine today on AI in RA, the briefing on artificial intelligence across research ecosystems. Here is the report. The Office of Research Integrity has stepped directly into artificial intelligence oversight, issuing standards for institutional research misconduct inquiries involving automated tools. It focuses on traditional violations—fabrication, falsification, and plagiarism—but compliance offices must now document the specific ways automated software contributed to the questioned data. That requires preserving prompt histories, running digital analysis on the outputs, and determining the boundary where legitimate software assistance ends and actionable fabrication begins. This guidance arrives as health agencies update their administrative ranks, including a proposed Chief Artificial Intelligence Officer at the National Institutes of Health and revised controls for generative software in regulated sectors. Universities are setting up risk-based frameworks to match those expectations. They are instituting baseline testing periods and requiring designated human accountability before software enters compliance workflows. The division of labor is deliberate: lower-risk administrative duties, like checking formatting before submission, can run on automated tracks, while institutional review boards and oversight committees retain manual authority over formal findings. Institutional records must withstand federal audits. That puts scrutiny on vendor data architecture, requiring proof that third-party platforms protect data privacy and produce legally defensible records during an investigation. That same barrier shows up in day-to-day research administration. Staff may use models to draft administrative language or checklists, but peer-review non-disclosure agreements and federal sponsor confidentiality rules prohibit uploading proprietary proposals or unfunded data into commercial engines. It creates a clear line between procedural support and actual proposal drafting. As covered in our June reporting on grant preparation, software can organize references or edit phrasing, but using it to generate substantive sections like needs statements risks introducing fabricated data points and stripping authorial voice. The agencies take different approaches to that boundary. The National Institutes of Health views proposals substantially produced by machine models as lacking original authorship, while the National Science Foundation permits assisted text provided the applicant submits an explicit disclosure. That federal clarity does not extend to the private sector. A Candid survey showed that a majority of philanthropic foundations have established no formal policies regarding applicant use of artificial intelligence. Leaving university development offices to track policies sponsor by sponsor. Campuses are attempting to organize that response from the top down; Rice University outlined an administrative framework during AI Week 2026 that balances operational automation with faculty oversight. Their design supplies central technology architecture and basic campus training, but leaves specific policy determinations for research administration and classroom work to individual departments. Turning to other policy and industry movements this cycle, the White House announced 2.4 billion dollars in compute pledges allocated to the Genesis Mission Consortium. Meanwhile, the Department of Homeland Security issued a proposed rule that would introduce fees for Optional Practical Training recommendations. Across enterprise adoption, Uber deployed ChatGPT Work to its entire workforce, and AT&T integrated ChatGPT to handle internal legal knowledge retrieval. To close the radar: OpenAI updated its storage architecture to accommodate over one billion users, the Allen Institute for AI published the LitPivot ideation tool, and federal policy documents shifted reference terminology from Artificial Intelligence to Super Intelligence. We will track how federal compliance standards continue to 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 →