Last updated: 17 October 2026, 04:23 BST.
AI research guidelines published by the European Commission on 16 October 2026 add guidance on increasingly autonomous agents and the assessment of research. For researchers in Sussex, the important distinction is that the European document does not replace university, journal or funder rules.
AI research guidelines: what is new?
The Commission’s announcement calls this the most comprehensive revision since the first guidelines in 2024. The previous revision was in May 2026. This update responds to systems that can plan and undertake longer sequences of tasks, rather than simply return one answer.
The fourth-edition document recommends human checkpoints for important methodological decisions. It also addresses disclosure, protection of unpublished material and limits on delegating scientific judgement in evaluation. These are recommendations, not a new permission to use an agent wherever a researcher chooses.
Why the Sussex distinction matters
The University of Sussex’s current information for postgraduate supervisors says supervisors and examiners must not use generative AI to review a postgraduate researcher’s work, upload any part of a thesis or associated work into such a tool, or use external AI-detection software on it.
That is a specific institutional restriction. A general recommendation about supporting tasks in European guidance is not an exception to it. Sussex’s thesis-submission guidance separately directs postgraduate researchers to its statement on acceptable use and appropriate referencing. It encourages discussion with supervisors.
What researchers should do before connecting a tool
Start with the task, not the product’s advertised capability. Identify whether you are drafting your own material, handling someone else’s unpublished work or participating in assessment. Find the rule for that activity before supplying any text or data.
For a permitted workflow, agree who checks the output, which decisions need approval and what record will make the work understandable later. Do not treat a plausible summary as proof that its source has been read accurately. Our separate report on Google’s laboratory-automation research examines the different problem of checking physical actions before a robot executes them.
Brighton Locals’ view
The useful advance is greater attention to what an agent does between instruction and result. But a broad ethics document cannot approve an individual workflow. For local research teams, checking the binding institutional rule and assigning a real human reviewer are more useful than simply attaching an “AI-assisted” label.
What to watch next
Look for institution- and funder-specific implementation. The European guidance is designed to evolve; that does not mean a Sussex policy has changed automatically or that a particular commercial tool has been approved.
Bottom line: newer AI guidance is a prompt to check responsibility and permissions, not to assume them.
Photograph: European Commission headquarters, Berlaymont, Brussels, 10 October 2026—not the guidelines launch or a Sussex laboratory. Foolxter / Wikimedia Commons, CC BY-SA 4.0; resized. Display crops retain the same licence.
Leave a Reply