Most students today understand that modern plagiarism checkers are so good that copying/pasting...
In the 11th issue of the flagship magazine CLOUD, published in February 2025 by the International Centre for Higher Education Innovation (ICHEI) under the auspices of UNESCO, Norvalid discusses “Academic integrity in the age of AI”. Our article discusses the challenges of AI's impact on academic integrity and explores effective methods for safeguarding assessments while preserving academic freedom and authenticity.
This blog post contains the full article. The original magazine can be found here.
AI offers fantastic opportunities for teaching and learning, but it also introduces risks to academic integrity. This article discusses how to safeguard assessments against the unethical use of AI, specifically large language models (LLMs) such as ChatGPT.
Universities are awarding bodies. Ensuring assessment security is essential to maintaining credibility. Assessing authentic student work is all that matters. The university's reputation and ability to uphold academic integrity define the value of the academic degree it awards.
Honesty and trust are core values of Academic Integrity.
According to the UNESCO IIEP blog: “Honesty, trust, fairness, respect, responsibility, and courage are fundamental values of academic integrity.” [1] In this context, honesty could be interpreted as students following instructions associated with their assignments and adhering to relevant university policies. Every assignment must include information on how students might leverage tools for assistance. Universities must address AI in their academic integrity policy. Student honesty can only be judged relative to what is expected of them.
The value of an academic degree relies on society trusting the university's ability to assess authentic student work. While everyone should be assumed to be honest, the problem arises if some gain an unfair advantage by cheating. Safeguarding assessment is essential for ensuring fairness and treating honest students with respect. Students are responsible for following policies and assignment instructions. Likewise, universities are responsible for assessment security to protect honest students and the value of the academic degree.
LLMs have broken academic integrity.
In October 2024, four Australian universities published a survey of 8,028 students, of which 41% answered that they “have used AI in assessment when I was not supposed to”, 71% “believe that AI increases cheating”, and 91% are “worried about breaking universities’ rules” [2]. 41% admit cheating. 91% are afraid of being accused of cheating.
While some are taking a wait-and-see approach concerning LLM cheating, these extraordinary numbers should be a wake-up call for senior management at all universities worldwide. Assessment security in the age of AI must urgently be addressed!
LLMs are challenging the “open book” home exam.
When discussing assessment security, we can distinguish between exams performed on campus and at home/remotely. Invigilated exams in managed environments on campus ensure assessment security, so LLMs should have little impact. The situation is different for home and remote examination. For onscreen exams, safeguarding is done through remote proctoring software, and for “open book” essay exams, safeguarding is done retrospectively by cheating detection software.
Remote proctoring involves controlling the students’ devices and surveilling the remote environment using a microphone, webcam, and sometimes a second camera. Since proctoring is intrusive, some argue that the security measurement limits the assessment, impacts student performance, and raises ethical concerns.
For “open book” essays, text similarity software detects plagiarism. However, it does not detect ghostwriting, and contract cheating is a problem. LLMs are the new world-class ghostwriters.
AI detection is not the solution.
A natural reaction is to try detecting LLM writing. While AI detection techniques might have a role to play, research finds that “available detection tools are neither accurate nor reliable” [3] and explain “simple techniques to bypass GenAI text detectors” [4]. There are also concerns about accusing honest students of cheating since “GPT detectors are biased against non-native English writers” [5]. According to this, detection is not a viable solution to LLM cheating!
Don't change the assessment. Critical thinking is more relevant than ever.
A few decades ago, plagiarism exploded when all humankind’s writing became available through the Web. The initial reaction was to change assessment due to plagiarism, but “open book” essays continued when text similarity detection software matured. Yet again, some say universities must change assessments due to LLM cheating. Maybe, but what is the pedagogical rationale?
“Open book” essays give students time and space to develop their thinking through writing. This is a fair way for all students to demonstrate their best ability to produce original work, justify a stand, or discuss ideas. An argument for essay writing being obsolete could be that students will use LLMs in their workplace. Counter-arguments could be the commonly accepted concept that “writing is thinking” and that academic degree holders must master critical thinking to judge output from LLMs.
The importance of challenging students to formulate their thinking in the age of LLMs is a pedagogic debate. It should not be mixed with the authenticity problem caused by LLM cheating.
LLMs are world-class ghostwriters.
Finding a solution to ensure authenticity starts by defining LLM cheating.
The “open book” essay allows the student to be assisted by any resource but requires uncited text to be original. With little student effort, LLMs produce excellent text, so it is tempting for some to claim they wrote it. Thereby, prompting LLMs is comparable to the age-old problem of ghostwriting—someone else is writing the essay, which is also the underlying problem of contract cheating, where the student pays a human author. By solving ghostwriting, we are solving both contract cheating and LLM cheating!
Ghostwriting is disclosed by school teachers every day.
Like LLMs, a human ghostwriter is hard to detect, yet school teachers solve the problem daily. Imagine you are a teacher following a class of 25ish pupils for some years. You know your pupils, so you get suspicious if an essay does not match the pupils’ writing style. Maybe the vocabulary, tone of voice, or the content not matching the pupils’ ability is catching your attention. You would naturally follow up with a conversation with the pupil to set doubts aside or confirm ghostwriting.
This method works well when the teacher knows the pupils’ writing and has the time to follow up. However, this is not feasible in higher education or for remote courses. But what if we can automate what a schoolteacher does to disclose ghostwriting?
Linguistic fingerprint to validate original student writing.
Few realize that all humans have an identifiable “linguistic fingerprint,” which can be extracted from an authentic writing sample of around 500 words. Students' unique linguistic fingerprints can be compared with their essays, judging whether it is by the same author as the writing sample. Relying on decades of research in linguistics and text forensics, the precision of these analyses can be 99%.
However, even a 1% false positive is problematic—adding a second test mitigates this. Students are asked questions about the content when they upload their essays to the learning platform. A custom-trained LLM formulates questions in real-time, which students must answer within a time limit as part of the submission process.
Using deep linguistic analysis to validate original student writing and asking questions about the content to check students’ knowledge gives substantial grounds for judging authenticity.
Authenticity in assessment is all that matters!
Only change assessments for pedagogical reasons. Universities should use prevention technologies to protect against technology cheating.
References
[1] Razi. S. Building an institutional culture of academic integrity. UNESCO IIEP ETICO Blog. (2020). https://etico.iiep.unesco.org/en/building-institutional-culture-academic-integrity
[2] Chung, J., Henderson, M., Pepperell, N., Slade, C., Liang, Y. Student perspectives on AI in Higher Education: Student Survey. Student Perspectives on AI in Higher Education Project. (2024). https://doi.org/10.26180/27915930
[3] Weber-Wulff, D., Anohina-Naumeca, A., Bjelobaba, S. et al. Testing of detection tools for AI-generated text. Int J Educ Integr 19, 26. (2023). https://doi.org/10.1007/s40979-023-00146-z
[4] Perkins, M., Roe, J., Vu, B.H. et al. Simple techniques to bypass GenAI text detectors: implications for inclusive education. Int J Educ Technol High Educ 21, 53. (2024). https://doi.org/10.1186/s41239-024-00487-w
[5] Weixin Liang, Mert Yuksekgonul, Yining Mao, Eric Wu, James Zou. GPT detectors are biased against non-native English writers. Patterns, Volume 4, Issue 7. (2023). https://doi.org/10.1016/j.patter.2023.100779
