One in five Dutch Master's Theses flagged as mostly AI-Written,

THE NETHERLANDS--An investigation into hundreds of master's theses submitted to 12 Dutch universities has found indications that approximately one in five was written predominantly using artificial intelligence, raising concerns about academic integrity, the credibility of university qualifications and the ability of educational institutions to determine whether students are producing their own academic work.

The findings, reported by Dutch broadcaster RTL Nieuws on October 14, 2026, follow an examination of recently completed theses using Pangram, a software tool designed to identify text potentially generated by AI chatbots such as ChatGPT, Claude and Gemini.

According to the investigation, approximately 6 percent of the examined theses were classified as almost entirely AI-generated, while another 14 percent contained more than 50 percent suspected AI-written content. An additional 39 percent contained smaller amounts of text flagged as AI-generated, while 41 percent showed no detectable AI-generated content.

The findings indicate that artificial intelligence may be playing a substantial role in the production of postgraduate research papers, although the results are based on automated detection rather than individual academic misconduct investigations.

AI detection software is not conclusive proof that a passage was produced by a chatbot. Such systems can incorrectly classify human-written material, and their findings generally require additional examination before a university can determine whether academic rules have been violated.

Nevertheless, the findings have prompted concern within the Dutch education sector about the extent to which students may be relying on AI systems to complete work intended to demonstrate independent research, analysis and academic writing.

Dutch Minister of Education Rianne Letschert characterized the use of AI to write a student's thesis as academic fraud and called on universities to strengthen their oversight.

She indicated that educational institutions must take greater responsibility for identifying inappropriate AI use and ensuring that qualifications reflect the knowledge and abilities of the students receiving them.

Letschert also announced plans to introduce a strategy addressing the issue before the end of 2026.

One of the more concerning aspects of the investigation involved differences between students' declarations about AI use and the findings generated by the detection software.

According to RTL Nieuws, almost half of the theses classified as predominantly AI-written included statements explaining how students had used artificial intelligence during their research and writing.

Many of these declarations described limited assistance, such as brainstorming, correcting grammar or spelling, and summarizing research material.

However, the software analysis suggested substantially greater involvement by AI tools in some of the completed documents.

At least two students reportedly declared that they had not used artificial intelligence, although their theses were classified as containing approximately 75 percent and 90 percent AI-generated material.

RTL Nieuws also interviewed four students whose theses were flagged as more than 80 percent AI-written. Some acknowledged extensive reliance on chatbots and suggested that such practices had become common among their peers.

One graduate reportedly used ChatGPT and Claude to produce much of a thesis, while another described using Claude for approximately 90 percent of the work.

These accounts raise questions about whether university policies are sufficiently clear and whether existing assessment procedures can reliably establish the extent of a student's independent contribution.

The investigation also found that the 12 universities examined, along with Universities of the Netherlands, did not maintain comprehensive figures on AI-related thesis fraud.

Policies governing the use of generative AI were also found to vary between universities and, in some cases, between individual courses.

Similar Concerns in the United States and England

The challenges identified in the Netherlands are not confined to Dutch universities. Higher education institutions in the United States and England are also reviewing how artificial intelligence is used in academic research, dissertations and postgraduate assessments.

In the United States, a study published in August 2026 in the International Journal for Educational Integrity examined AI policies at 50 universities recognized for innovation. It found that institutions were adopting different approaches, including restrictions on unauthorized AI-generated work, requirements for students to disclose AI assistance and changes to assessment methods.

Some universities have adopted particularly restrictive policies. The University of Georgia, for example, prohibits the use of generative AI in theses and dissertations unless specifically authorized. Other American institutions permit certain uses of AI while requiring students to acknowledge the assistance received.

The research also identified concerns about the reliability of AI detection software, with universities increasingly considering methods that evaluate students' research processes and understanding rather than relying exclusively on automated checks.

In England, the University of Birmingham has established guidelines prohibiting postgraduate students from submitting AI-generated dissertation content unless specifically permitted by their academic department.

The university allows limited AI-assisted proofreading under defined conditions but requires disclosure and places restrictions on using the technology to develop arguments, alter ideas or produce substantive academic content.

Birmingham has also recommended that supervisors monitor students' progress throughout the dissertation process by reviewing drafts, conducting regular meetings and examining how research arguments develop. The intention is to ensure that students can demonstrate ownership of their work.

These developments illustrate a growing international challenge for universities: determining where legitimate technological assistance ends and academic misconduct begins.

Questions About the Value of Academic Qualifications

The increasing ability of AI chatbots to produce research summaries, formulate arguments, generate references and draft extended academic texts is changing how students approach postgraduate assignments.

A master's thesis has traditionally been intended to demonstrate that a student can independently investigate a subject, evaluate evidence, develop arguments and present conclusions supported by research.

When substantial portions of that work are generated by a chatbot, universities face the question of whether the completed document accurately reflects the student's own understanding and academic abilities.

There are also concerns about accuracy. Generative AI systems can produce convincing text containing factual errors, unsupported claims or references to sources that do not exist. Students who rely on these systems without checking their output risk submitting inaccurate material.

At the same time, universities recognize that artificial intelligence can provide legitimate educational benefits, including assistance with language, research organization and understanding complex subjects.

The challenge is therefore not necessarily to prohibit the technology entirely, but to establish rules that allow appropriate assistance without replacing the intellectual work students are expected to perform.

The Dutch investigation highlights the difficulty of enforcing those boundaries as AI-generated writing becomes increasingly difficult to identify through conventional academic assessment.

For universities in the Netherlands, the United States, England and elsewhere, the broader concern is maintaining confidence that a master's degree represents the knowledge, research skills and independent judgment of the individual who earned it.

With the Dutch government preparing a national response before the end of 2026, the findings are likely to contribute to wider discussions about academic standards and how universities should assess students in an education system where sophisticated AI writing tools are readily available.

‍