Three University of Crete doctoral regulations published in the Greek government gazette on 28 August 2026 make undisclosed AI use a removable offence and prescribe the exact wording a candidate must use to declare it

The quality floor, and where it sits

In the Political Science regulation the AI rules are not in a technology annex. They are in Article 12, titled plagiarism and the use of Artificial Intelligence systems, which places AI misuse in the same provision as plagiarism and attaches the same consequence.

Article 12.4 sets the substantive test. It reasons from the nature of doctoral work: a thesis must constitute an original contribution to knowledge, and therefore must by definition exceed what can be produced by means of Artificial Intelligence. The use of AI systems and applications is not justified in substituting for human thought. Verbatim copying or reproduction of content from AI systems is prohibited and may lead to the removal of the doctoral candidate, constituting a serious academic offence.

That formulation is worth noticing because it defines the standard by reference to the capability of the tools rather than by listing forbidden products. A rule written that way moves as the tools move, which is either its main virtue or its main problem depending on who has to apply it.

Article 12.3 adds the procedural leg: alongside the thesis, the candidate must file with the three-member advisory committee a plagiarism-check certificate reporting the result of a check run with the University of Crete Library's text-similarity application, named in the gazette as Turnitin.

The model disclosure sentence

The regulations go further than the prohibition and supply the words. Where AI systems or applications have been used, the candidate's declaration in the thesis is required to take a prescribed form, printed in the gazette with ellipses where the tool name goes. It appears in both the Political Science and the Preschool Education regulations in the same terms.

Rendered in English, it reads: for the processing of the text of this doctoral thesis the Artificial Intelligence tool [name] was used exclusively for language editing, clarification of theoretical concepts and structural prototyping. The final text was checked, modified and approved by its author. All scientific analyses, judgements and citations are based on primary personal research.

Two things follow from putting that sentence in a gazette. It fixes the permitted purposes inside the disclosure itself, so a candidate who used a tool for anything else cannot truthfully sign it. And it gives examiners a fixed string to look for rather than an open-ended judgement about whether disclosure was adequate.

The Economics regulation carries the same idea into the sworn declaration a candidate files, offering a branching text: either an affirmation that no AI tools were used in conducting the research or in processing the text, or, if they were, the same enumerated-purpose disclosure.

Which document says what is worth setting out plainly, because the three are not copies of each other. The Political Science regulation is the fullest: it alone carries the Article 12 heading, the exceed-what-AI-can-produce test and the similarity-check certificate, and it also contains the disclosure sentence, the non-disclosure offence and the exhaustive permitted-use list. Preschool Education carries the disclosure sentence, the non-disclosure offence and a statement that AI is permitted only in a supporting role, but not the quality floor or the certificate. Economics carries the permitted-use list and the sworn declaration.

Non-disclosure is the offence

This is the provision most likely to be missed, and it is the one that gives the regime its teeth. The Political Science and Preschool Education regulations both state that supportive use of AI systems and applications without this being expressly declared in the thesis, and without the manner of their use being clearly described, is itself a serious academic offence capable of leading to removal of the candidate.

So there are two distinct routes to removal. One is using AI to do the work, through verbatim reproduction without critical human processing. The other is using AI permissibly and then failing to say so, or saying so without describing how. A candidate who stays comfortably inside the permitted purposes can still be removed for silence about it.

Both place responsibility for the result on the candidate: a candidate who uses AI systems or applications bears responsibility for the accuracy and the legality of the final content of the thesis. Legality, not merely accuracy, is doing work in that sentence.

A closed list, not a standard

The Economics and Political Science regulations are the explicit ones about scope, in identical terms. During the conduct of doctoral research and for processing the text of the thesis, the use of AI tools is permitted exclusively for identifying sources, language editing, clarification of theoretical concepts and structural prototyping. All texts, scientific analyses, judgements and citations must be based on the candidate's own primary research, for which the candidate is solely responsible.

That is drafted as an exhaustive list rather than as an open standard with examples, and the word carrying the weight translates as exclusively. Anything not on the list is outside the permission, which is a materially different drafting choice from the more common approach of prohibiting substitution and leaving permitted uses implicit.

For anyone building an institutional policy, the three documents together are a usable template: a capability-referenced quality floor, an exhaustive permitted-use list, a prescribed disclosure string, a similarity-check deliverable, and a named sanction with an existing procedural route. They are also, usefully, short.

What we did not verify

We fetched all four gazette issues flagged to us and read them ourselves. FEK B 5321, 5322 and 5323 extracted cleanly with no replacement characters, and every provision described above was located in the extracted text and mapped clause by clause to the specific issues that contain it. The three are not identical texts, and we have said which carries which rather than treating them as one document.

One correction to the research that reached us. A fourth issue, FEK B 5284 of 26 August 2026, a University of West Attica internal operating regulation, was described to us as part of this cluster. It is not. We searched the whole 685,780-character document and it contains no reference to artificial intelligence in any form. Its 31 replacement characters sit inside a single 131-character ECTS credit formula, nowhere near any relevant text, and three independent extraction modes agree. We have excluded it, and this article is about three regulations from one university, not four from two.

We did not open Law 4957/2022, under which these Senate regulations are made, and we have not independently confirmed the scope of that enabling power. We did not open the University of Crete Postgraduate and Doctoral Studies Regulation, so articles 36 and 42 are described only as the route the gazette names, not as text we have read.

We did not verify whether any candidate has been removed under these provisions, and we make no claim about how they are applied in practice. We also make no claim that these rules extend beyond the three departments named, or beyond the University of Crete.

Related briefings

• Greece: the national AI Act framework, HDPA and EETT
• Germany: OVG NRW on generating exam answers by smartphone
• Spain: AEPD on biometric facial recognition in exams
• Europe AI regulation hub
• AI Regulation News hub