Institutions escalate, students adapt, and learning drops out of the exchange
Well-intentioned institutional policies can turn educational relationships adversarial, with each new control met by a new workaround. Seeing the conditions that set this cycle going explains why responses to AI that rely on prohibition and detection tend to fail.
Arms race dynamics in higher education
One-sentence definition: A pattern in which institutions escalate controls on students, students adapt around them, and the cycle displaces the educational goals both sides started with, driven by misaligned incentives, trust deficits, and fixation on measurable proxies.
An arms race is a competition in which each move provokes a counter-move, so everyone involved keeps spending more just to hold their position. Harland et al. (2015) described an assessment arms race in the grading practices of a New Zealand university. Every grade counted towards the degree, students worked only on what was graded, and so a lecturer who wanted work done had to grade it; as colleagues graded more, everyone else followed, until one module carried 31 graded tasks. Nobody wanted the result, and no one could step out of it alone, while learning became secondary to collecting marks. Harland and Wald (2021) later traced how it built up over several decades. The same structural pattern now appears in institutional responses to AI tools, in prohibition without purpose, detection-led policy, and punitive frameworks, and students answer them with increasingly sophisticated ways of evading detection.
Conditions that create educational arms races
Three structural conditions transform educational relationships into adversarial cycles:
Misaligned incentives: parties optimise for different outcomes. In grade proliferation, academics wanted engagement while students wanted to protect their marks; neither primarily optimised for learning. In AI prohibition, institutions want authenticity verification while students want efficiency and support; learning sits awkwardly between them.
Trust deficits: institutional assumptions create self-fulfilling prophecies. In the grade arms race, lecturers came to expect that students wouldn’t work without grades, and the system kept that expectation true. The AI arms race assumes students will cheat if they can. Both assumptions shape policies that make the feared behaviour more likely.
Measurement fixation: complex educational interactions get reduced to quantifiable proxies. Grades proxy for engagement; detection software proxies for authentic work. The proxy becomes the target, displacing what it was meant to measure.
Reflective writing on placement shows all three at once. A paramedic science programme that runs students’ placement reflections through an AI detector signals that it expects them to cheat, rewards whatever passes the detector, and measures authenticity by a score. Students learn that the risk lies in sounding too polished and can start writing flatter, more guarded reflections to stay clear of a flag. The detector may catch some generated text, but the programme has traded away the candour that made reflection worth assessing.
Limits of the metaphor
Arms races imply mutual escalation, but in the AI case students are adapting to escalation the institution drives. The institution introduces detection software and students find methods it can’t detect; it raises the penalties and students get better at evading them.
That distinction matters for intervention. If it’s truly mutual escalation, you need to de-escalate both sides. If it’s institutional behaviour creating resistance, the intervention point is institutional policy.
The question it changes
Arms race dynamics are most visible in contexts where institutions attempt to control student behaviour through prohibition and surveillance rather than through aligned incentives and transparent rationale. High-stakes assessment environments, technology adoption periods, and academic integrity policies all create conditions where these dynamics can emerge. Recognition matters because it shifts focus from “how do we catch rule-breakers?” to “what about our system creates rule-breaking as a logical response?” Underneath all of this sits the imbalance of power between institutions and students.
Sources
- Harland, T., McLean, A., Wass, R., Miller, E., & Sim, K. N. (2015). An assessment arms race and its fallout: High-stakes grading and the case for slow scholarship. Assessment & Evaluation in Higher Education, 40(4), 528–541. https://doi.org/10.1080/02602938.2014.931927
- Harland, T., & Wald, N. (2021). The assessment arms race and the evolution of a university’s assessment practices. Assessment & Evaluation in Higher Education, 46(1), 105–117. https://doi.org/10.1080/02602938.2020.1745753