Let us get one thing out of the way: if a company promises you a "100% AI-proof" exam at the click of a button, they are selling you something they cannot deliver.
The technology moves too fast, and the ways students use it are too varied, for any single software feature to be a magic shield. We know this. We think you deserve to know it too, before you buy anything from anyone.
Why the silver bullet does not exist
Here is what is happening in the industry right now: companies are selling AI-detection tools as the answer to AI-generated answers. It sounds logical until you try to use one in a real classroom.
Detection software is playing a game it cannot win. These tools produce false positives, flagging honest students who happen to write in a clear, plain style. They produce false negatives too, missing students who paraphrase or who use whatever came out last week and is not in the detection database yet.
It is an exhausting arms race. More to the point, it puts you in the position of policing rather than teaching.
Some platforms go the other way and double down on invasive proctoring: eye tracking, room scans, continuous video. That approach assumes distrust is the foundation of assessment. It creates anxiety, it raises real privacy problems, and it still does not guarantee integrity. A determined student with a second device or a well-placed note works around it.
We are not going to sell you that dream.
The north star and the reality
So where does that leave us?
Our position, stated plainly: a 100% AI-proof exam experience is a north star. It is the direction we build toward and the standard we measure features against. It is not a claim about today.
In practice it means something more honest and more useful: continuously reducing the avenues for unauthorised assistance, improving how quickly a problem is detected and answered, and being transparent about the limits of any technology across the enormous variety of devices and rooms students actually sit exams in.
See the difference? We are not claiming to have arrived. We are telling you where we are headed and how we are getting there, and we are being straight that technology has limits which shift depending on the device a student is holding, their connection and their environment.
That is not a weakness. It is the only foundation trust can be built on.
Layered defence is the only defence that works
If there is no magic button, what actually works?
Layered defence. Several barriers that together make cheating both harder to do and less worth doing.
Layer one: technical restrictions
The exam runtime removes the obvious shortcuts. Being precise about what each one does matters more than making it sound impressive:
- Focus locking. Leave the exam screen, even once, and the attempt locks. A browser cannot stop a student opening another tab, and any vendor who tells you theirs does is describing something that does not exist. What it can do is notice immediately and end the attempt, which is what this one does.
- Copy blocking on the question text. A deterrent rather than a wall: the words are on the screen, so they can be retyped or photographed. It removes the effortless path, not every path.
- Server-side grading. The answer key never reaches the device. There is nothing in the page to read, whatever the student opens the developer tools on.
- One attempt per email address, so an exam cannot simply be reloaded for a second go.
None of this is foolproof, and a determined student with a second device can work around all of it. But most students are not determined cheats hunting for elaborate workarounds. They are taking the path of least resistance. Layer one removes that path.
Layer two: assessment design
This is where most platforms stop. Technical restrictions on their own are incomplete, which is why they have to be paired with better questions.
When you write exams that ask students to apply knowledge to a specific context, analyse a scenario they were in the room for, or solve a problem particular to your course, you are doing something more effective than any lock: you are making AI assistance less useful.
"Explain the water cycle" is answered by any model in seconds. "Analyse how the water cycle would be disrupted in the catchment we surveyed in week three, and propose two interventions" needs context, synthesis and applied judgement that a model cannot fake convincingly without having been there.
It bears repeating: the best defence against AI is not better detection. It is better questions.
Layer three: transparency
The third layer gets overlooked: telling students the rules in plain language.
- What is restricted during the exam
- What is monitored, and why
- How their privacy is protected
- What happens if something unusual is detected
When you treat students as parties to the integrity of the exam rather than suspects to be caught, compliance goes up and anxiety goes down. The contract becomes legible: the platform protects the exam, the educator uses the controls responsibly, and the student knows the conditions before they begin.
What this means for you
Be sceptical of any platform promising complete protection through a single feature. Whether it is billed as AI detection or AI-proofing, if it sounds simple, it is not true.
Treat integrity as a system, not a switch. It is the combination of technical controls, question design and clear expectations that makes an assessment credible.
And know what you are buying. We make it meaningfully harder to cheat, we support the kind of applied questions that reduce what AI is worth in the first place, and we tell you what our tools can and cannot do.
Where this leaves us
Can we make an exam that is completely immune to AI assistance? Not today, not with current technology, and not across the infinite variety of devices and circumstances your students turn up with.
Can we make it significantly harder to cheat while making the assessment itself more meaningful? Yes. That is the work.
And can we be honest with you about where the technology is and what it can realistically do? That is what this is.
Trust is not built on promises nobody can keep.