The 2 Keys Every School Needs to Address to Reduce AI Related Academic Misconduct
When we ask the 3,000+ schools we work with globally what their biggest AI challenge is, the answer is almost always the same: academic misuse. Students using AI to draft and submit assignments and assessments for them. When we ask how they're going about solving the problem, the answer is generally one of two things.
• We've moved solely to exams.
• We've tightened our rules.
Unfortunately, there are a few problems with these approaches. Firstly, a world in which the only form of assessment is pen and paper is a pretty bleak one. An exam only environment has real repercussions for student engagement and satisfaction. Secondly, it deprives students of voice and agency in their learning, a step backwards after years of work to make learning more relevant and student led. Finally, it doesn't facilitate the very thing we're all aiming for, which is student learning. Exams may let us measure what students have temporarily learned, but not the long term effects of learning or genuine knowledge acquisition.
The real question is how schools keep interesting, meaningful assessment without giving up academic integrity. We've found schools have two levers: building AI literacy and rebuilding assessment. Get these two right and you address both why students misuse AI, and how much opportunity they have to do it.
Lever 1: Build AI Literacy Skills
The mistake many schools make when they reflexively fall back on exams is assuming all misuse comes from malicious intent. In reality, much of it comes from students simply not understanding how AI, and specifically Large Language Models (LLMs) like ChatGPT, Gemini and Claude, actually work.
Take a story an art teacher at a client school told us recently. He had set students an assignment analysing an impressionist artist and how they diverged from traditional art forms. While marking, he came across an artist he had never heard of. Certain it was a spelling mistake, he checked, then searched the name. No such artist existed. The work was invented.
But something didn't sit right. It didn't read like a student who had made up a random name to dodge the work. It was in the student's own voice, with their usual writing quirks. The next day, he gently raised the issue of academic misconduct. The student, genuinely shocked, insisted they had written the whole thing themselves and done the research. So the teacher asked what they had used for research. The answer: a well known LLM.
This wasn't a student setting out to cheat. It was a student who simply didn't understand the technology they were using, and they were making three of the most common mistakes students make with AI.
Misunderstanding the technology. Students often treat LLMs like Google or a super database, somewhere you input a query and receive a factual answer back. That's a basic misunderstanding of how an LLM works. It has no knowledge in itself. It has simply been trained on a gigantic volume of words, which it uses to determine, probabilistically, what word comes next. As Ethan Mollick puts it, in the simplest terms, an LLM is essentially autocomplete on steroids. Understanding this is a basic AI literacy skill, and it's what leads students to be appropriately sceptical of any AI output.
Confusing confidence for fact. Students mistake an LLM's confidence for knowledge. It sounds certain, and it's usually right, so it starts to feel like a knowledgeable authority, almost like a parent or a teacher. But an LLM is a consumer product, built to keep users subscribing and returning. Its makers, OpenAI, Anthropic, Google, Microsoft, know that users don't want a tool that voices self doubt or hedges its answers. So LLMs are built to sound confident and state assertions as fact, even when they shouldn't be. Understanding this mechanic is another key AI literacy skill, one that leads students to view any AI output with a critical eye.
Not verifying outputs. Given the two points above, every student needs to be taught how to verify and validate an AI output. This means learning to pause before using an output, being able to identify quotes, facts, statistics, names, events and articles within it, and knowing where to go to validate and verify them so they can be confident in their accuracy. Of the three skills, this is arguably the most transferable, since it's exactly the kind of critical thinking students will need long after they leave school.
Schools looking to reduce misuse should start by building an inventory of these core AI skills, an AI literacy curriculum. And these skills need to be taught across year levels, not just Years 11 and 12. Over the last 12 months, we've noticed most capability building for students concentrates on senior years, which makes sense given schools want to ensure students are using AI ethically as they graduate. However, schools are increasingly finding their Year 7 students arrive already exposed to AI and using it regularly. A comprehensive program should target core skills in Year 7 and continue building on them all the way through to graduation. For an example of a scaffolded AI skills inventory, click here.
Lever 2: Rebuild Assessment
Building AI literacy helps with unintentional misuse, but it doesn't necessarily help with intentional misuse. Policies and rules help here, and so, as mentioned, does moving to pen and paper exams. But rules alone won't eliminate misuse, and moving solely to exams is neither desirable nor sustainable. The reality is that the biggest opportunity schools have for reducing misuse is redesigning assessment itself.
Yet we've found only 30% of schools have actually started that work. We understand why. Reworking every assessment is extremely time consuming, and it's easier to postpone than to start. More importantly though, most teachers simply don't have the knowledge or confidence to work through the process. In most staff rooms, we've found attitudes toward AI are normally distributed. At one end, a small group of teachers already using it extensively and enthusiastically. At the other, a small group who are avowedly anti AI and don't believe it has any place in education. In the middle, a large group who are curious, interested, or want to experiment, but don't feel confident enough to know where to start. They may not see themselves as technology people, or find the jargon and technical talk too complex and intimidating. These teachers have the will, but not yet the way.
As we'll cover in an upcoming newsletter, we believe assessment redesign is actually the best place to start building teachers' AI capability and confidence. Start with professional development that helps teachers:
• Audit their assessments to determine how easily an LLM could shortcut the task
• Understand the characteristics of assignments that make this possible
• Understand the principles of AI resilient design
• Decide what level of AI access they want to allow students
• Build disclosure processes that measure and reward good AI use, rather than just act as a reporting mechanism
The beauty of this approach is that it doesn't just make assessments harder to complete with AI, reducing misuse cases. It also makes assessments more pedagogically sound. It gets teachers interacting with AI in a number of different ways, which quickly builds breadth of experience, and these multiple interactions help increase teachers' confidence.
Finally, this kind of PD should be backed by the school setting some general principles on what a good AI balance looks like. Many schools we work with now use the AI Assessment Scale developed by Leon Furze, which categorises AI use across 5 levels: Level 1, no AI; Level 2, AI used for planning, idea development and research; Level 3, AI used collaboratively for drafting and refining; Level 4, full AI, used extensively as directed; and Level 5, AI exploration, creative and undirected use of AI to approach the problem. This scale is an effective way for teachers to communicate to students how much AI is allowed, and even more importantly, it helps a school think through what balance is right for them. Schools that use the scale well tend to deliberately calibrate different assignments to different levels, rather than defaulting to one level across the board. It gives teachers clear guidance that they can't simply default to Level 1 because it's easy, while giving the school a clear structure for how much PD and capability building is needed to get the balance right. As a deputy principal told us recently, their goal is to get every teacher capable and confident setting assignments at Level 3. Right now, only half their staff are there.
For schools currently struggling with AI misuse, these two levers, building AI literacy and rebuilding assessment, provide the perfect starting point to move beyond simply relying on rules and pen and paper exams.
Want to see exactly where your school sits on assessment readiness and AI misuse compared to the schools we've benchmarked? Try a free module of our AI Culture Snapshot: here. It's a quick way to see, in the same terms used throughout this piece, exactly where your school's biggest opportunities sit.
Comments