EU research guidance draws a line: AI can assist review, not make the judgement
The fourth edition adds agentic AI, stronger disclosure expectations and tighter peer-review safeguards. It is a non-binding translation of existing integrity principles, not a new law.
The short version
- What happened
- Researchers remain accountable when agents perform sequences of tasks and should validate critical methodological decisions at explicit checkpoints.
- Who it affects
- Researchers gain a clearer disclosure threshold, while reviewers and applicants receive a stronger confidentiality safeguard for unpublished work.
- What remains unproven
- The guidance supplies principles rather than one implementation standard and presents no outcome data showing that the recommendations reduce misconduct or bias.
Keep in mind
The guidance supplies principles rather than one implementation standard and presents no outcome data showing that the recommendations reduce misconduct or bias.
Source analysis · 2 linked sources. Primary analysed source date: 16 October 2026.
Read the full report · 4 min The evidence, methods, implications and limitations
Autonomy does not transfer responsibility
Researchers cannot treat an AI agent as the party responsible for a method, analysis or conclusion, according to the European Commission's updated living guidance. The fourth edition adds explicit instructions for agentic systems: researchers should define checkpoints, validate critical methodological decisions and remain accountable for every phase in which an agent is allowed to plan or act.
The update also sharpens meaningful disclosure. If AI substantially shapes a literature review, research gap, hypothesis, data analysis or interpretation, users should identify the system and explain its contribution and human verification. Routine spelling, formatting or minor language polishing will not usually need the same disclosure. Researchers are asked to consider retaining prompts, logs, model versions and parameters when they materially affect reproducibility.[1][2]
Peer review gets the clearest boundary
The document says scientific-quality and merit judgements must not be delegated to AI. Reviewers, editors and funding evaluators may use permitted support functions, but they remain responsible for assessments and recommendations. Unpublished manuscripts, proposals, review reports and ideas should not be pasted into unsecured public AI systems; where AI is allowed, the guidance prefers secure tools with clear governance and data management.
Funding organisations are told to watch for biased or erroneous screening, ranking, prioritisation and exclusion. Institutions are encouraged to provide training, track AI use, monitor environmental impact and, where feasible, offer locally hosted or governed systems for sensitive data. Researchers are reminded that uploading personal data requires a lawful basis and that an agent acting on their behalf does not remove legal obligations.
This does not mean every cloud tool is prohibited. The guidance asks users to examine who owns and operates the system, where it runs, what happens to prompts and uploaded material, and whether data can be reused for training. That makes procurement and information governance part of research integrity, rather than leaving each researcher to infer privacy from a consumer settings page.[1][2]
The distinction that matters
Where the updated guidance places responsibility
Requirements become stricter as AI moves from minor assistance to judgement or autonomous action.
Minor support
Spellcheck, format or polish language
Usually no substantive-use disclosure
Research method
Review literature, analyse data or run an agent
Disclose material use and document verification
Evaluation
Judge a paper or funding proposal
Scientific merit remains a human decision
It is a common reference, not an enforcement regime
For a doctoral researcher, supervisor or reviewer, the practical test is clearer: could another person understand where AI materially influenced the work, and did a human retain control over confidential information and consequential judgement? Institutions can turn that into disclosure fields, approved-tool lists and review checkpoints without banning every low-risk use in practice.
The authority check matters. This European Research Area stakeholders' document is explicitly non-binding. It translates existing integrity and trustworthy-AI principles; it creates no new penalty and does not override the AI Act, data-protection law, funder terms, journal policies or national rules. Disciplines may adapt the recommendations proportionately. What would change the assessment is implementation evidence: comparable disclosure, fewer confidentiality breaches, reproducible AI-assisted methods and audits showing that automated screening does not skew funding. Until then, the update supplies a stronger governance baseline, not proof that European research has solved the problem.[1][2]
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 research integrity: compare the finding and its evidence limits.
Can AI tools review scientific literature reliably? A 102-paper benchmark found access and judgment gaps
Five tools collectively found 57 of 102 known papers before a controlled subscription test. Given every PDF, three tools found all 18 central papers but misclassified 40% to 79% of passing mentions as substantial. The manuscript is an unreviewed preprint.
Read the next reportExplore the full source and evidence record
What this means for people
- Researchers gain a clearer disclosure threshold, while reviewers and applicants receive a stronger confidentiality safeguard for unpublished work.
Global context
The document is designed for the European Research Area, but journals, funders and collaborators operate internationally. Cross-border projects may face additional rules.
What the evidence does not yet show
- The guidance supplies principles rather than one implementation standard and presents no outcome data showing that the recommendations reduce misconduct or bias.
What to watch next
- Institutional policies, secure research-AI services, disclosure templates, funder audits and evidence about bias or confidentiality incidents in AI-assisted evaluation.
Living evidence record
Impact record IAI-1DPTYEM
Evidence stage
Announced
Confidence
Supported
Reporting basis
Source analysis
Independent or research support
Not yet
Record status
Monitoring
Last checked
16 October 2026
Source trail
2 direct sources across 2 source types.
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.
Related-source reporting disclosure
This record analyses 2 linked source records around the same underlying development. The extra records add method, date or context, but they do not by themselves constitute independent replication of every performance claim or predicted outcome.
Evidence trail
Sources used for this report
Links checked 16 October 2026
Research question, topics and regions
Topics: Research integrity · AI agents · Peer review · Funding · Confidentiality
Regions: European Union · Europe
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.