Students and lecturers in Leiden draw up 32 rules for fair assessment in the AI era

At Leiden University, 47 students and lecturers from all seven faculties sat down together this spring to do something that few institutions have dared to attempt so far. Together, they devised concrete rules for fair assessment in an era where generative AI is creeping in everywhere.

The initiative is called FAIR-ASSESS and consisted of four half-day sessions, held between January and February 2026. The reason is a problem now recognised by every school and university. Students use AI during exams and assignments, but lecturers are also increasingly deploying it to design assessments, provide feedback, and even determine grades. This raises questions about validity, transparency, and whether such practices are actually permitted under the European AI Regulation.

32 rules, none with full agreement

The four sessions resulted in 32 proposed rules, built on eight principles: validity, reliability, transparency, usability, trust, authenticity, accountability, and equality. Strikingly, not one of the 32 rules achieved complete unanimity. The report itself refers to them as rules "for which there was moderate to substantial support", not rules on which everyone agrees.

This lack of full consensus is precisely what makes this initiative interesting. It shows that even within a single university, there are genuine differences of opinion between disciplines and between students and lecturers. Having that conversation, rather than imposing a top-down, watertight policy document, is what we believe will truly help schools and universities move forward.

What this means for EduGPT

At EduGPT, we recognise the core of what Leiden is grappling with. Anyone deploying AI in assessment must be clear in advance about the purpose. Does it support the lecturer in marking, or does it take over part of the marking process? That distinction is exactly why we always keep the lecturer in control at EduGPT. AI provides feedback and identifies where a pupil is struggling, but the lecturer makes the decisions and does the marking.

Transparency, too—one of the eight principles from the Leiden report—is built into how we have designed EduGPT as standard. Pupils and lecturers can see when and how AI has been used, rather than this remaining hidden in a black box.

Especially in an age of AI, it is reassuring to see a university like Leiden conducting this discussion seriously and methodically, with students at the table rather than just lecturers and administrators. The full advice of the Deliberative Assembly can be read via Leiden University.

Photo: USFWS/Southeast, CC BY 2.0

Video clip on assessment and AI

CivAI has produced its own knowledge clip on this subject, "Assessment, Learning and AI – From Choices to Responsibility". The clip shows the choices schools have when deploying AI in assessment and where responsibility then lies, a fine complement to the Leiden initiative above.