A head of department told me recently that her school had stopped setting take-home essays for Year 11 coursework components altogether, not because the syllabus changed, but because she no longer believed the essays that came back to her were an honest record of what any individual student could do.
Most teachers I talk to describe something similar. As one put it to me: "It's a cat and mouse game. The kids know that they're better than us at technology, and simply take the shortest route to homework. The fact that the 'shortcut' is sometimes more effort than actually doing the work doesn't seem to bother them at all. The chance to brag about finishing quickly, presumably with the least time and effort, beats any 'real learning' ideal. It's exhausting, this isn't what we signed up for as teachers. Redesign? Sure, but we need help."
That "we need help" is the part worth sitting with, because it is not a complaint about workload in general, it is a specific request: teachers can see the redesign coming, and most have neither the time nor the mandate to do it department by department, alone, on top of everything else already on their plate. That is not a small decision for a school running IGCSE. Coursework components exist precisely because some things, sustained argument, independent research, drafting and redrafting, cannot be tested well in a two-hour exam hall. Withdraw take-home assessment because you no longer trust it, and you have not solved the integrity problem: you have narrowed what you are able to assess at all, which is its own kind of loss.
The exam hall was never the whole answer
The instinct to retreat to supervised, in-person assessment is understandable, and for a portion of the marks, it is the right instinct: nothing currently substitutes for watching a student produce work in real time, under known conditions, with no device in reach. But IGCSE coursework was built the way it was because supervised exams measure a narrower set of skills than a full subject deserves, and a school that quietly reduces coursework's weight to sidestep the AI problem is trading a solvable integrity issue for a permanent curriculum one.
The more useful question is not "exam hall or take-home," but which specific tasks still work as take-home assessment now that a capable model can draft convincingly on almost any prompt, and which need to be redesigned or moved.
Before the next coursework cycle, have each department list its take-home assessment tasks and mark each one "AI-resistant," "AI-vulnerable," or "unsure." A task is AI-resistant if it depends on something specific to that student's process, in-class notes, a source only discussed in a particular lesson, an annotated draft, rather than only on the final product. This single afternoon of triage tells you more about where your actual exposure sits than any policy document will.
What actually distinguishes a defensible task
Tasks that survive contact with a language model tend to share a feature: they require evidence of process, not just a finished product. A structured draft-and-feedback sequence, where the teacher sees and comments on an early outline, a mid-draft, and a final version, is far harder to fake convincingly than a single essay submitted cold, because it requires sustained, specific engagement with the teacher's actual comments over several weeks, not a one-off generation. Vivas and short oral defences of written work, even five minutes per student, do the same work: a student who can discuss why they made a particular argumentative choice has almost certainly engaged with the material, whether or not AI assisted along the way, and one who cannot, reveals that fast.
None of this requires new software or a detection tool, which is worth saying plainly, because detection tools are the wrong layer to solve this at. They generate false positives against strong writers whose natural style happens to resemble model output, false negatives against students who lightly edit AI drafts, and an arms race that trains students in evasion rather than honesty. The schools I have spoken with that are furthest ahead on this treat detection as, at best, a weak secondary signal, and put their actual weight on redesigning the task so the question of "did AI write this" becomes less important than "can this student demonstrate they understand it."
Why This Matters for IGCSE Schools Specifically
Cambridge International and Pearson Edexcel both permit AI use in some coursework contexts provided it is declared and the student's own understanding is demonstrable, which puts the burden of proof on the school's assessment design, not on a blanket ban that is unenforceable in practice. A policy that simply forbids AI use, without redesigning the tasks that made the prohibition necessary, asks teachers to police something they cannot reliably detect, and asks honest students to take on faith that dishonest classmates aren't gaining an advantage. Malaysian international schools sitting IGCSE have the same coursework components, the same integrity exposure, and, I would argue, less institutional memory of having navigated a comparable disruption before, having built assessment practice in a relatively stable period. That is not a criticism; it simply means the redesign work has to happen deliberately now rather than being inherited from an earlier crisis.
Want to work through this with your assessment leads?
I run short discovery conversations with heads of department and academic leadership teams on where coursework tasks are genuinely exposed, and what a practical redesign looks like for your specific subjects, not a generic policy.
Book a Discovery CallWhat This Means for You
The schools that come through this well will not be the ones with the strictest AI ban, or the most sophisticated detection software. They will be the ones that looked task by task at what they are actually asking students to prove, and redesigned the minority of tasks that no longer prove it. That is unglamorous, department-by-department work, and it is also the only version of this that actually holds up once the next model generation arrives and makes today's workarounds look quaint.