This summer, the highest education court in the Netherlands set a number of benchmarks for cases involving AI. In three extensive rulings, the Council of State explains when the use of AI is and is not acceptable.

For AI-related fraud, the judge makes clear, there must be solid evidence. One striking detail: in one case, the education judge actually Googled the student’s thesis to check it. That rarely happens. Normally, the judge mainly looks at whether the rules have been followed correctly.

“Carelessness”

The first case concerns a marketing student at the University of Groningen who submitted a thesis containing AI-generated errors in the reference list. The student immediately admitted this to the examination board. He had an explanation: afterward, he had ChatGPT reorganize his reference list. That was when the errors crept in.

Was this really fraud? The student himself called it “carelessness.” He submitted a corrected reference list. And he could show that an earlier version of the bibliography contained no hallucinations, meaning he had not relied on AI for the actual content.

The examination board and, later, the Examination Appeals Board (CBE) nevertheless found fraud. The main text of the thesis contained three references to sources that were missing from the bibliography. Those, too, must have been AI fabrications, they concluded.

Do your own research!

The judge decided to check that for himself. With the thesis in hand, he sat down at a computer to track down the three sources. He succeeded. As the judge writes, “after a simple search,” he was able to find a relevant academic source for all three references, by the author named and from the correct year.

The three references in the main text were therefore not AI hallucinations. According to the judge, the student did indeed have a plausible explanation: ChatGPT had simply made a mess of his bibliography.

Not every AI hallucination is fraud, the judge therefore ruled in this case. Sometimes it is simply a mistake. And that is not grounds for punishment by the examination board, but rather for a lower grade from the instructor assessing the work, the Council of State told the program.

Examination boards and Examination Appeals Boards spend a great deal of time checking suspected cases of fraud, as, for example, examination boards in Maastricht told university newspaper Observant. But in this case, the judge found that the examination board and the CBE at the University of Groningen had not done enough. The student’s explanation “should have been reason” for them to do a quick Google search.

Forgot the AI statement

Was the student even allowed to use AI to structure his bibliography? That question was not addressed in the Groningen case. In the Rotterdam case, it is at the heart of the matter. The master’s program in Media & Business at Erasmus University has very specific rules on this, as the second ruling makes clear.

Students were only allowed to use AI if they also submitted an AI statement. This had to specify what students had asked ChatGPT (the “prompts”) and what they had incorporated into their work. With the AI statement, the instructor could still assess the students’ own knowledge and skills.

The student who brought the case had not submitted an AI statement for the first assignment and had submitted an incomplete one for the second. It was clear that she had used AI to compile her bibliographies. ChatGPT had made up quite a few things in the process. She said she had done it because she was short on time. According to her, the main text of her papers was accurate.

And that is the crux of the matter: this Rotterdam student had AI generate the entire bibliography, whereas the Groningen student had only asked it to reorganize his. That is an important distinction for the judge. The bibliography is a “full-fledged component” of the submitted student work, the judge explains. She loses her case.

Sufficient evidence?

A student from Maastricht denies before the judge that she used AI at all. For her European Law program, she had to submit a short literature review every week, and the examination board claimed that she had used AI for three of her four assignments. There were conspicuous errors in years, titles, page numbers and journal titles.

Before the judge, she offered an alternative explanation: she was short on time and had a poor internet connection, forcing her to rely on her memory and incomplete notes. That led to errors. On top of that, she has difficulties such as dyslexia, which make reading, writing and spelling difficult for her.

To support her case, she submitted her old notes so that the Council of State could see for itself how the errors in her assignment had occurred. But the judge did not find those errors in her notes. There was no conclusive evidence, but to the judge, “fraud” was simply the best explanation. She therefore did not prevail.

The same principle applies to AI as to other fraud cases: it comes down to weighing the evidence and investigating possible alternative explanations. AI-related fraud must be established “beyond reasonable doubt,” the Council of State ruled in these three cases. That standard, “beyond reasonable doubt,” may explain why relatively few AI cases have so far made it to court.

This article was translated using AI-assisted tools and reviewed by an editor.