AI agents are no longer just a topic for software companies. They are finding their way into laboratories, libraries and research offices. On 16 October, the European Commission published an updated version of the ERA Living Guidelines on the responsible use of generative AI in research — the most comprehensive update since the guidelines were first released in 2024.

For the first time, the guidelines include specific guidance on agentic AI: tools that can plan and carry out complex tasks with a degree of independence. They also give much more detail on using AI in peer review and in the evaluation of funding proposals. For Irish universities, research institutes and funders, the update offers a practical framework for using these tools with confidence.

What the updated guidelines say

The guidelines are voluntary recommendations for researchers, research organisations and funding bodies across the European Research Area. They are built on familiar research-integrity principles: accountability, transparency and responsibility. The update keeps the simple, practical style of earlier versions, with a revised structure that makes the guidance easier to follow.

The previous revision came in May 2026. The Commission says this latest version responds to rapid advances in generative AI and to feedback from the scientific community, gathered through an open process involving countries and stakeholders across Europe.

Where agents can help Irish researchers

There is plenty of everyday research work where AI agents can save real time. Literature searches, coding, cleaning datasets, drafting summaries and preparing presentations of findings are all areas the guidelines recognise as legitimate support.

The key recommendation is to build in checkpoints. When an agent is working through a multi-step task, researchers should stop at sensible points to review and validate its decisions — especially anything that affects methodology. The responsibility for scientific outputs stays with the researcher, even when an autonomous tool did much of the work.

For Irish research groups, this is a workable model. A PhD student could use an agent to scan hundreds of papers and propose a shortlist, then check the shortlist personally before it shapes a study design. A lab could let an agent draft analysis code, with a human signing off before results are reported.

Where human judgement stays

The guidelines draw a firm line around research assessment. Expert judgement in peer review and grant evaluation must not be handed over to AI tools. Scientific assessment remains a human responsibility.

Confidentiality is the second pillar. Unpublished proposals, manuscripts and review reports must not be shared through unsecured AI tools. Reviewers are also expected to follow the specific rules set by the funder or journal they are working for.

That gives Irish funders and universities a clear starting point for their own policies. Research offices can set out which approved, secure tools staff may use, train reviewers on what is and is not allowed, and make disclosure of AI use a routine part of applications and papers.

What Irish institutions can do now

Many Irish universities already have generative AI guidance for teaching and learning. The next step is research-specific guidance that reflects agents: approved tools, clear checkpoints, rules for handling confidential material and simple ways to record how AI was used in a project.

Because the guidelines are designed as a living document with an open feedback process, Irish researchers also have a chance to shape future versions by sharing what works in practice.

The Bottom Line

Europe's updated research guidelines give Irish universities a practical green light to use AI agents for the heavy lifting of research, as long as people stay in charge of the decisions that matter. Agents can speed up searches, coding and drafting; humans keep ownership of methods, results and every peer-review or funding judgement. Institutions that turn these principles into clear local rules now will help their researchers move faster without putting trust at risk.