Do research funders have clear rules for AI use? Only 12 of 46 did
A peer-reviewed document analysis found that only 12 of 46 selected funding bodies directly addressed AI use in grant applications or review. The sample is a purposive, website-visible set rather than a global census, but it exposes a practical governance gap for applicants, reviewers and institutions.
The short version
- What happened
- The researcher began with 117 funders listed by eight publicly funded Hong Kong universities, excluded 71 that did not meet the criteria, and analysed 56 AI-related documents from 46 bodies across six regional groupings.
- Who it affects
- Researchers may face uncertainty about permitted drafting help, disclosure and confidential material when a funder publishes only broad principles or no visible rule.
- What remains unproven
- The sampling frame came from funding calendars at eight Hong Kong universities and is not a representative census of global funding bodies.
Keep in mind
The sampling frame came from funding calendars at eight Hong Kong universities and is not a representative census of global funding bodies.
Regional group sizes were highly uneven, including only two bodies for Europe and two for Mainland China and Hong Kong.
The website search was updated in May 2026; internal, poorly indexed, later or differently worded policies may be absent.
Document counts do not measure implementation, compliance, enforcement, applicant understanding or effects on funding decisions.
The thematic categories involve interpretation despite strong reported inter-coder agreement.
The study does not establish that publishing a policy improves research integrity or produces fairer outcomes.
Source analysis · 1 linked source. Primary analysed source date: 5 November 2026.
Read the full report · 9 min The evidence, methods, implications and limitations
The answer: clear funding-specific rules were the exception in this selected group
For most of the 46 research funding bodies in this study, a researcher or reviewer could not find a public policy written specifically for AI use in funding. The author identified 56 AI-related policies and guidance documents, but only 13 were funding-specific. Just 12 bodies had rules that directly addressed AI in an application or assessment process. That is a policy-availability finding, not proof that the other organisations ignore AI internally, and the denominator is a selected set rather than every funder worldwide.
The practical distinction matters. A broad national AI principle can say systems should be fair, safe or accountable without telling an applicant whether a language model may edit a proposal, whether generated text must be declared, whether confidential applications may be placed in an external tool, or how a reviewer should treat suspected use. Funding-specific guidance can convert general values into decisions that affect eligibility, confidentiality and peer review. When that translation is absent, applicants and reviewers may improvise different rules for the same competition.
The paper, published in the peer-reviewed International Journal of Educational Technology in Higher Education on 5 November, is a systematic document analysis. It does not measure misconduct, grant quality or whether a policy changes behaviour. Its contribution is narrower and useful: it shows what an academic researcher could locate on official funder websites, how those documents were coded and where stated principles stopped short of operational instructions. This is current policy research, not an urgent event, so the story belongs in Latest rather than Breaking.[1]
How the 46 funders were selected and what the researcher counted
The starting list contained 117 funding bodies drawn from research-service calendars at Hong Kong's eight publicly funded universities. Bodies were included if they provided or administered external research funding available to academics, belonged to one of six geographical groupings and had an identifiable public official website. Duplicate entries were consolidated and 71 organisations were excluded, leaving 46. The regional counts in the analysed set were 24 in the United States, eight in Australia, six in the United Kingdom, four in Singapore, two in Europe and two in Mainland China and Hong Kong.
Researchers manually examined official websites in March 2025 and updated the search in May 2026. They checked navigation and subpages and used site searches for terms including AI policy and AI guideline. The resulting corpus contained both general organisational AI documents and materials aimed at funding applications or review. Combining them let the study ask whether high-level organisational commitments were being converted into instructions at the point where money is awarded.
This design is transparent but purposive. A funding body absent from the eight university calendars could not enter the frame, and an internal or poorly indexed rule could be missed. Europe is represented by only two bodies, while the United States contributes more than half the sample. The study therefore cannot support a claim that a particular percentage of all global funders has a policy. Its denominator is 46 bodies visible and relevant through one institutional starting point, and every percentage should remain tied to that group.[1]
The distinction that matters
From funder lists to a policy corpus
The denominator narrows at each step, so the 12 bodies with direct guidance should be read against the selected 46, not all global funders.
117 listed bodies
Apply access, region and public-site criteria
46 funders retained
46 funders
Search official sites through May 2026
56 AI-related documents
56 documents
Separate general from funding-specific rules
13 funding-specific documents
Application and review
Identify direct operational guidance
12 of 46 bodies
What the coding found beyond the headline count
Inductive thematic analysis produced five themes: human-centredness, safeguarding, embeddedness, governance and AI readiness. Human-centredness covers human oversight and responsibility. Safeguarding includes privacy, security, fairness and related protections. Embeddedness concerns integrating AI rules with existing research and institutional processes. Governance addresses accountability and oversight structures, while AI readiness concerns the capacity, literacy and resources needed to apply the rules.
The gap was not simply that some organisations had no document. Funding-specific materials tended to address immediate application and review questions but, according to the paper, did not contain the governance and AI-readiness themes found elsewhere in the corpus. AI readiness appeared in only 16 of the 46 documents overall. That suggests a rule can prohibit disclosure of confidential material or require human responsibility without explaining who trains reviewers, audits compliance, handles appeals or updates practice as tools change.
To test coding reliability, a second researcher independently reviewed all 312 coded extracts. The two coders agreed on 288 and disagreed on 24, an agreement rate of 92.3%; Cohen's kappa was 0.90. Disagreements were resolved through discussion, with the sole author retaining final responsibility. This is stronger than unreviewed single-coder interpretation, but it does not make the themes objective measurements. Policy language can fit more than one category, and the coding framework does not determine whether the policy works in practice.[1]
Regional absence in the sample is a prompt for checking, not a league table
The paper reports that none of the selected funders from Singapore or Mainland China and Hong Kong had a funding-specific AI policy. The 12 bodies with direct application or review guidance included organisations in the UK, US, Australia and Europe. That regional contrast is consequential for cross-border collaborations: partners may face different disclosure duties even when contributing to the same proposal. It should not be read as evidence that one region is generally less responsible, because only four Singapore bodies and two bodies in Mainland China and Hong Kong were included.
Public visibility also reflects language, web architecture and institutional form. The study searched official websites and analysed documents that could be found publicly. A national ministry, a mission-oriented charity and a research council may publish rules through different channels. A later policy could also have appeared after the May 2026 search update. The correct use of the regional result is to ask each funder for a current, authoritative rule and to record the date and scope, not to infer a country's entire governance capacity from a small purposive subset.
For universities, the immediate task is compatibility checking. Research offices can map each major funder's rules against institutional requirements for proposal drafting, confidential material, authorship, disclosure and review. Where a funder is silent, institutions should say which local policy applies and how applicants can obtain a written answer. Reviewers also need guidance: they should not paste unpublished proposals into consumer AI tools merely because a public policy is missing, and they need a route to disclose and manage any permitted assistance.[1]
What the study does—and does not—show about research integrity
A missing public policy does not establish harmful AI use, and a published policy does not establish compliance. The study did not survey applicants or reviewers, examine grant files, audit decisions, compare acceptance rates or test whether disclosure changes trust. It also did not score policies against a legally validated standard. The count is evidence of uneven public rule-making, while claims about consequences remain a policy interpretation that needs behavioural data.
The paper's funding should be visible when judging independence. It reports support from the University of Hong Kong Seed Funding Programme and the Research Grants Council of Hong Kong General Research Fund, grant 17616525. The author declared no competing interests and reported assistance with website identification, extraction and independent coding. No dataset was generated or analysed in the conventional sense; the evidence is the public document corpus described in the paper and supplementary material.
The assessment would change with a representative registry of funders, multilingual searches, archived policy versions and direct confirmation from organisations about unpublished or newly issued rules. Surveys and audits could then test whether applicants understand requirements, reviewers handle confidential material safely and comparable cases receive comparable treatment. Until then, the responsible conclusion is that operational funding guidance was uncommon in this selected public corpus, not that global research funding has no AI governance.[1]
Keep following the evidence
Join The Impact Brief founding list for selected AI reports, their effect on people and the evidence limits. Read a sample first.
Continue the story
Read next
Continue exploring ai governance: compare the finding and its evidence limits.
What will the new US AI task force actually decide?
President Donald Trump has publicly announced a federal AI task force led by intelligence chief Jay Clayton. The announcement confirms its four-person leadership and stakeholder remit, but the charter, evidence process, enforcement power and final policy test remain unpublished.
Read the next reportExplore the full source and evidence record
What this means for people
- Researchers may face uncertainty about permitted drafting help, disclosure and confidential material when a funder publishes only broad principles or no visible rule.
- Peer reviewers need explicit limits on using external AI services with unpublished proposals, plus training and an escalation route for ambiguous cases.
- University research offices can reduce unequal treatment by keeping dated, competition-specific guidance and obtaining written clarification when funder rules are silent.
Global context
Funding rules travel through international collaborations: one proposal can involve researchers, data and reviewers in several jurisdictions while being governed by a single funder's terms. The study spans selected bodies in North America, Europe, Australia and Asia, but its regional denominators are too uneven for country rankings. A useful next step is a multilingual, versioned registry that distinguishes general AI principles from operational rules for applications, peer review and funded research.
What the evidence does not yet show
- The sampling frame came from funding calendars at eight Hong Kong universities and is not a representative census of global funding bodies.
- Regional group sizes were highly uneven, including only two bodies for Europe and two for Mainland China and Hong Kong.
- The website search was updated in May 2026; internal, poorly indexed, later or differently worded policies may be absent.
- Document counts do not measure implementation, compliance, enforcement, applicant understanding or effects on funding decisions.
- The thematic categories involve interpretation despite strong reported inter-coder agreement.
- The study does not establish that publishing a policy improves research integrity or produces fairer outcomes.
What to watch next
- Representative, multilingual mapping of funders with archived policy dates and direct organisational confirmation.
- Clear disclosure, confidentiality, authorship and reviewer-use rules attached to individual funding calls.
- Evidence that applicants and reviewers understand the guidance and that comparable cases receive comparable treatment.
- Governance mechanisms for training, appeals, audits and policy updates rather than prohibitions alone.
Living evidence record
Impact record IAI-0VBQ07B
Evidence stage
Studied
Confidence
Supported
Reporting basis
Source analysis
Independent or research support
Present
Record status
Monitoring
Last checked
5 November 2026
Source trail
1 direct source across 1 source type.
People impact
Documented in this record.
Uncertainty
Limits and next checks are explicit.
Stages describe the evidence available—not whether a technology is good or bad. See the public method.
Single-source reporting disclosure
This record analyses one direct source. It can establish what International Journal of Educational Technology in Higher Education published or reported, but it is not independent corroboration of every performance claim or predicted outcome. The confidence label will change only when broader evidence is added.
Evidence trail
Sources used for this report
Links checked 5 November 2026
Research question, topics and regions
How selected public research funders translate broad AI principles into rules for grant applications, assessment and funded work
Topics: Research funding · AI governance · Grant review · Research integrity · Policy implementation
Regions: Asia · Europe · North America · Australia
This report is labelled source analysis. We summarise and analyse source material in our own words; company statements remain attributed claims until independently supported. Translated summaries preserve the meaning of the original source and link back to it. Read our editorial standards.
Reader commentary
What does this mean where you are?
Tell us what you have seen at work, in your community or in everyday life. A question is welcome too. No account is needed; we review notes before publishing them.
Explore commentary across the portal →Published reader notes
0No published reader notes yet. You can start the evidence-led discussion above.
Prefer a private correction or response? Contact the newsroom.