“Where’s the line?” If you’ve taught a course in the past two years, you’ve probably heard that question from students. You may have asked it yourself. Where does acceptable AI use end and academic misconduct begin? And who decides?
Corbin, Dawson, Nicola-Richmond, and Partridge (2025) took that question and studied it empirically. Their paper, published in Assessment & Evaluation in Higher Education, reports on interviews with 19 students and 12 teachers at a large Australian university. Nobody prompted the “line” metaphor. Participants reached for it unprompted, again and again, in over half of all interviews. And what they described wasn’t a line at all. It was a blur. As one teacher put it: “It makes absolute no sense. Where’s the line? It’s an absurd line” (p. 708).
What makes this paper valuable is that it gets past the institutional framing and into the actual lived experience of the people navigating this confusion daily. A large number of current AI assessment literature talks about what policies should look like. Corbin et al. talk about what the absence of clear policy feels like.
Students Are Building Their Own Rules
One of the strongest findings is that students aren’t waiting around for guidance. Faced with vague or nonexistent institutional policies, many have constructed their own ethical frameworks for AI use. Some decided AI is fine for understanding concepts but off-limits for producing final text. Others treat it like Grammarly: acceptable for polishing, not for generating.
One student reasoned through it almost philosophically:
I guess interacting with the Chatbot it does make me think about collusion, because of the way that the bot speaks as if it is sentient. It speaks like it has sentience. And so, it feels like collusion to talk to it. Which is not true, because it’s not actually expressing any thoughts or anything like that, or its own opinions but yeah, but because it’s collusion, I kind of treat how I interact with it as if I was interacting with another person. So, if I wouldn’t ask another person to write my essay, why would I ask this bot to? […] So yeah, so I guess that’s kind of how I structured the boundaries I have with it.(p. 710)
These self-made frameworks are thoughtful. They’re also fragile, because they rest almost entirely on guessing what the teacher believes. As one student described the situation: “You have to kinda guess your teacher’s perspective on AI, and then figure out how much to use it or not” (p. 708). That’s a high-stakes guessing game. Students are making decisions about academic conduct based on inference, not instruction, and several described genuine fear of being accused of cheating for something they considered reasonable.
I covered a related dynamic when writing about the European Commission’s JRC study on AI use in secondary schools across five EU countries (Villar Onrubia et al., 2025). The finding there was similar: students were already using AI extensively, and their teachers were largely unaware of how or how much. Corbin et al. add the emotional layer. The confusion creates anxiety that falls unevenly. Students who are less confident or less familiar with institutional norms carry more of the risk.
And students themselves know that banning AI won’t work. One said plainly that a flat-out no-AI policy is “just a pipe dream” (p. 712). Another described their university’s position as essentially: “we can’t stop you, but we don’t recommend it” (p. 712). That kind of ambiguity leaves the boundary-drawing entirely to students, which is exactly where it shouldn’t be.
Teachers Are Bearing the Weight Alone
The teacher side of the data is just as striking, but for different reasons. Several participants described emotional strain connected to marking. One said it directly: “I’ve been highly anxious all year marking. I have been second guessing myself” (p. 713). Another described the experience as “hugely hurtful” (p. 711). A third, who had taken long service leave, wondered aloud whether they even wanted to return to a system they called “a bloody sweatshop” (p. 712).
Some teachers had redesigned their assessments. Others felt paralyzed. One admitted: “I’ve done absolutely nothing in terms of assessment practices because I don’t have a clue” (p. 709). And notably, the teachers who had taken action often did so entirely on their own, without institutional resources or recognition. One described the personal cost explicitly: “although I enjoyed it and I’ve got a bit of notoriety from it, it’s bloody killed me” (p. 711).
But not all teachers felt helpless. One of the more interesting quotes in the paper comes from a teacher who reframed the problem entirely: “if my exam can be answered by an AI, if it’s an open book exam, I have not done my due diligence as an educator to tailor my exam to the unique knowledge of the unit” (p. 712).
I found that perspective refreshing. It shifts responsibility from policing students to improving assessment design, and that aligns with the broader argument I’ve been developing across this blog. The question shouldn’t be “how do we stop students from using AI?” The question should be “why does our assessment make AI use so easy and so tempting?”
Corbin, Bearman, Boud, and Dawson (2025), in their earlier paper on the wicked problem of AI and assessment, proposed three “permissions” institutions should extend to educators: permission to compromise, permission to diverge, and permission to iterate. The data in this newer paper shows why those permissions matter so urgently. Without institutional backing, teachers absorb all the uncertainty, and the toll is real.
Why the Line Keeps Moving
Corbin et al. ground their analysis in boundary work theory, drawing on Gieryn (1983) and Lamont and Molnár (2002). Boundary work describes how communities construct and maintain the borders between legitimate and illegitimate practice during periods of disruption.
AI has triggered exactly that kind of disruption. Older categories like plagiarism or contract cheating assumed a clear substitution of student work. AI doesn’t work that way. It blends into every stage: brainstorming, outlining, drafting, editing, even comprehension. The boundary between “I used AI to help me understand” and “AI wrote this for me” is genuinely hard to draw, and participants on both sides of the desk struggled with it.
Both students and staff tried to stabilize the boundary by reaching for analogous technologies. Grammarly came up repeatedly. The reasoning was simple: if Grammarly can rephrase your sentence and that’s fine, why is ChatGPT different? The analogy has limits, but the impulse makes sense. When the rules are unclear, people look for precedent.
The AI Assessment Scale from Perkins, Roe, and Furze (2024) helps here because it replaces the vague “line” with five explicit levels of AI integration, each tied to specific learning goals. But even with a scale, teachers still need to decide which level fits which task. The boundary is still negotiated. It’s just negotiated with more clarity.
The Framework They Propose
Corbin et al. offer what they call the Dynamic Educational Boundaries Model (pp. 714-715), built on three dimensions.
The first, assessment-embedded boundaries, argues that expectations about AI use should be built into the task structure. An oral defense inherently limits AI outsourcing. A task that explicitly requires documented AI interaction reframes use as purposeful. I agree with this completely. Design the assessment around the thinking you want students to do, and much of the ambiguity dissolves. Hartmann (2025), whose paper on oral exams I covered recently, did exactly this: she restructured an entire course so the assessment format itself made the expectations clear.
The second, contextual flexibility, recognizes that appropriate AI use varies by discipline and task. A statistics course and a marketing project raise very different questions. Universal bans miss this. So do blanket permissions. As one teacher in the study noted, “each course would have to have their own guidelines based on how they’re going to actually use it” (p. 715). I agree with that too.
The third dimension, emotional and cognitive support, is the weakest part of the model. Not because the idea is wrong. Teachers absolutely need professional development, shared repositories of examples, and peer mentoring. But the paper doesn’t specify what that support looks like in practice. “Provide professional development” appears in nearly every paper on AI in education.
The question is always: what kind? How deep? For how long? Bilbao-Eraña and Arroyo-Sagasta (2025) showed in their study on AI literacy for teachers that eight hours of training shifted awareness and attitudes but didn’t build trust. Support that stays vague stays ineffective.
Where Does This Leave Us?
Corbin et al. close with a useful framing: “This study reveals how the seemingly simple question of ‘where’s the line?’ in AI use masks profound challenges to educational practice and identity” (p. 715). That’s accurate. The line question sounds like it should have a simple answer. The fact that it doesn’t frustrates everyone involved.
But frustration isn’t a reason to stop working on it. The teachers and students in this study are already doing the hard labor of figuring out where AI belongs in academic life, without enough support, clarity, or trust. Institutions owe them better. Clear frameworks. Ongoing professional development. Assessments designed so expectations are visible from the start. The line will always move. The goal isn’t to fix it permanently. The goal is to make sure everyone knows where it is right now, and why.
References
- Bilbao-Eraña, A., & Arroyo-Sagasta, A. (2025). Fostering AI literacy in pre-service teachers: Impact of a training intervention on awareness, attitude and trust in AI. Frontiers in Education, 10, 1668078.https://doi.org/10.3389/feduc.2025.1668078
- Corbin, T., Dawson, P., Nicola-Richmond, K., & Partridge, H. (2025). ‘Where’s the line? It’s an absurd line’: Towards a framework for acceptable uses of AI in assessment. Assessment & Evaluation in Higher Education, 50(5), 705-717. https://doi.org/10.1080/02602938.2025.2456207
- Corbin, T., Bearman, M., Boud, D., & Dawson, P. (2025). The wicked problem of AI and assessment. Assessment & Evaluation in Higher Education. Advance online publication. https://doi.org/10.1080/02602938.2025.2553340
- Gieryn, T. F. 1983. “Boundary-Work and the Demarcation of Science from Non-Science: Strains and Interests in Professional Ideologies of Scientists.” American Sociological Review 48 (6): 781–795. doi:10.2307/2095325.
- Hartmann, C. (2025). Oral exams for a generative AI world: Managing concerns and logistics for undergraduate humanities instruction. College Teaching. Advance online publication. https://doi.org/10.1080/87567555.2025.2558563
- Lamont, M., and v. Molnár. 2002. “The Study of Boundaries in the Social Sciences.” Annual Review of Sociology 28 (1): 167–195. doi:10.1146/annurev.soc.28.110601.141107.
- Perkins, M., Roe, J., & Furze, L. (2024). The AI Assessment Scale revisited: A framework for educational assessment (Preprint). December 2024. https://arxiv.org/abs/2412.09029
- Villar Onrubia, D., Cachia, R., Rietz, C., Feltrero, R., Niemi, H., Hallissy, M., & Reuter, R. (2025). Generative artificial intelligence in secondary education: Uses and perceptions from the perspective of early adopters across five EU Member States. Publications Office of the European Union. https://publications.jrc.ec.europa.eu/repository/handle/JRC144345