On 16 October 2026, the European Commission announced its most comprehensive update yet to the ERA Living Guidelines on the responsible use of Generative AI in research. The revision addresses increasingly autonomous AI tools and expands guidance on peer review and funding-proposal evaluation.
Developed with countries and stakeholders across the European Research Area, the guidelines offer a common reference for researchers, universities and funding organisations. First published in 2024 and previously revised in May 2026, they translate research-integrity principles into practical guidance as AI capabilities and research practices evolve.
What has changed?
The update brings agentic AI explicitly into the discussion and clarifies earlier advice on evaluating proposals. These developments raise questions about who supervises increasingly complex tasks, who remains accountable and how the reliability of AI-assisted work is established. The revised structure keeps practical guidance, transparency and responsibility at its centre.
Agentic systems can plan and carry out sequences of tasks with some autonomy. The guidelines call for human checkpoints, including validation of methodological decisions. Researchers remain responsible throughout the work: giving an agent a task does not transfer control or accountability to the system.
Human judgement remains central to evaluation
Reviewers must follow the relevant funder, journal and institutional policies. Where permitted, AI may support limited tasks such as clarifying concepts or language polishing. It must not assess scientific merit or draft evaluation reports. Reviewers retain responsibility for assessments and decisions, and must protect unpublished proposals, manuscripts and review material from disclosure through unsecured tools.
When should AI use be disclosed?
The guidance distinguishes substantive contributions, such as literature reviews, hypothesis development or interpreting results, from minor editing. Substantive use should be disclosed, explaining the tool's role and verification. Basic spelling, grammar or formatting support generally does not require disclosure, subject to applicable policies.
Protect sensitive work and support staff
Researchers should avoid uploading sensitive, unpublished or protected material unless suitable safeguards are in place, and check the tool's privacy settings. They remain responsible for the accuracy and integrity of AI-assisted work. Training and sharing good practice are also part of responsible use, particularly as tools and their capabilities change.
For organisations, the recommendations extend beyond individual behaviour. Universities and research institutes are encouraged to include AI in internal policies, involve staff in shaping those policies and provide secure AI environments under organisational governance wherever possible. Funding bodies should give evaluators clear guidance and help applicants explain substantial AI contributions to proposals or planned research.
What this means in practice
For the research community, the update connects everyday AI use with existing expectations of trustworthy research:
- Research teams: retain responsibility for the work, protect sensitive inputs and explain meaningful AI contributions.
- Universities and institutes: combine clear policies with training and access to appropriately governed tools.
- Funding organisations: make permitted uses clear, support disclosure and maintain human responsibility in funding processes.
The guidelines are non-binding. They complement existing law and institutional, disciplinary and funder requirements; they do not replace them. Specific applicable rules take precedence.