Lighthouse AdvisorySLED AI Adoption Intelligence
← Back to results

From the Research edition of September 10, 2026

Standards or public-body guidanceCautionaryNewly relevant · Dec 2023

NSF guidance distinguishes proposal assistance from confidential merit review

U.S. National Science Foundation · University research administration and federal research funding · United States; NSF merit review

Publisher
NSF
Original publication
December 14, 2023
Source retrieved
2026-09-11
Read original source

What happened

NSF prohibits reviewers from uploading proposal and review content to non-approved generative AI tools.

Why it matters

Directly relevant to university researchers serving as NSF reviewers; this is not a universal rule for every sponsor.

Evidence and measured results

The notice encourages disclosure of proposal-development AI use and retains proposer responsibility for accuracy and authenticity. It supplies policy requirements, not measured workload or compliance outcomes.

Limitations and uncertainty

Foundational 2023 notice re-inspected today, not a new 2026 prohibition. No efficacy sample or baseline. This edition does not resolve every sponsor's current requirements.

Put this evidence to work

Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-09-11; this does not change the original publication date. Labels below come from the analysis itself.

Sales

Role takeaway

Speak with sponsored-programs leadership, research security staff and faculty reviewers about confusion between writing support and assessment duties. Ask which sponsor governs each task, what information staff would upload, and who can authorize its use. Offer a bounded workflow and permissions assessment using public or synthetic examples. The value hypothesis is clearer routing and fewer inappropriate disclosures, not proven administrative savings. Do not position private hosting, a vendor contract or a campus license as permission to process review materials. Keep the engagement scoped to verified sponsor requirements and avoid promising automatic grant evaluation.

Pre-sales engineering

Role takeaway

Design distinct workspaces for authorized proposal preparation and restricted reviewer activity. Inventory browser extensions, document connectors, agent tools, logs and outbound model calls so hidden routes do not defeat a nominal boundary. Use public test documents to verify that restricted workflows cannot invoke unapproved endpoints. Apply least privilege and record the source of each permission decision. Proposed validation should include denied transfers and readable error messages as well as allowed tasks. The notice establishes an authorization constraint, not an implementation recipe; confirm any deployment design with the responsible review official before using confidential material.

Delivery

Role takeaway

A sponsored-programs owner should maintain the workflow inventory with research security and faculty representatives. Dependencies include identified sponsors, current approval records and clear escalation contacts. Train staff on their role for each task, including researchers who alternate between applicant and reviewer duties. Gate rollout on documented authorization and test the controls again after connector changes. Proposed acceptance criteria include an owner for every workflow, no successful restricted-data transfer in negative tests, and an accessible manual alternative. Measure support burden and exceptions. Risks include role confusion, stale guidance and users assuming tool availability means permission.

Implementation considerations

Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.

Architecture and integration

What must connect, and where does the AI sit in the workflow?

Separate proposer assistance from reviewer workspaces and data paths.

Governance

Who approves, reviews and stays accountable for outcomes?

Confirm sponsor-specific authorization before approving a review workflow.

Security and privacy

What data, permissions and controls need testing?

A campus enterprise subscription alone does not establish NSF approval.

Accessibility and workforce

Who is affected, and what skills or accommodations follow?

Provide accessible approved workflows and an escalation route rather than assuming an assistive tool is authorized.

Procurement

What should contracts, pricing and exit terms secure?

Verify the allowed use and data boundary before evaluating licensing or security assurances.

Operating model

Which teams own the service once it runs?

Sponsored-programs staff maintain a sponsor/workflow matrix; review officials resolve authorization questions.

What changed

No matching policy URL or title in the full archive. Newly covered control for research-administration assistance; older date is explicit and no policy change is claimed.

Publication history

  1. 2026-09-10Research · Issue 053 resources
Read preserved resource versions (JSON)

Stable resource ID: nsf-merit-review-genai-boundaries-2023