AI is Changing Education. Assessment now has to catch up

The debate about whether AI belongs in education is rapidly becoming redundant. The more important question is what education does now that AI is already here and BETT’s latest research makes that shift difficult to ignore.

Weekly AI use among teachers has doubled in a year, from 26% to 52%. And its reach extends far beyond regular users: only 11% of teachers say they have never used AI for a teaching purpose. Yet the same technology helping teachers create lesson resources, draft communications and complete administrative tasks is creating a much harder question when it appears on the other side of the classroom: how do we know what a student genuinely knows?

That question goes far beyond detecting which AI tool a student may or not have used.

The problem isn’t simply AI cheating

Concern about academic integrity is understandable. Bett found that 71% of secondary teachers believe AI makes cheating easier, while 57% had suspected a pupil of submitting unauthorised AI-assisted work within the previous four weeks.

The instinctive response is to look for better ways to catch it. More monitoring. Better detection. Tighter restrictions.

But that risks turning assessment into a technological arms race.

As students gain access to increasingly sophisticated AI tools, institutions may find themselves turning to equally sophisticated, and increasingly intrusive, forms of monitoring to detect their use.

There will be circumstances where strong controls are entirely appropriate. But before deciding how closely a student needs to be monitored, educators need to answer a more fundamental question: What is this assessment actually designed to prove?

That sounds obvious. AI has made the answer much more important.

Not every assessment should work in the same way. There will always be high-stakes assessments where students need to demonstrate foundational knowledge independently.

If an exam is designed to establish whether somebody possesses a particular body of knowledge without external assistance, then access to AI, search engines or other resources fundamentally changes what is being tested. Strong controls make sense.

However, that isn’t the purpose of every assessment.

Many should establish whether students can analyse information, apply knowledge to unfamiliar situations, evaluate evidence and explain the reasoning behind their conclusions.

In those cases, allowing access to certain resources does not automatically invalidate the assessment. Depending on what is being tested, controlled access to online material – and potentially AI – could form part of it.

After all, education is preparing people for workplaces in which digital tools will increasingly be part of how tasks are completed.

The challenge is ensuring the assessment measures the student’s capabilities rather than the tool’s.

That could mean asking students to explain how they reached a conclusion, defend their decisions or demonstrate the steps behind their work. A correct answer tells us something. Understanding why somebody believes it is correct can tell us considerably more.

Design out the advantage of cheating

Interestingly, teachers themselves appear to recognise that assessment needs to change.

When Bett asked what should happen to written coursework in response to AI, 30% favoured keeping coursework while adding better AI detection. However, 29% preferred replacing it with supervised in-class writing and another 9% wanted extended-writing coursework removed altogether. Only 4% wanted to leave things as they are. Those findings should prompt a broader conversation about designing assessments in which cheating provides less of an advantage in the first place.

If an assignment can be completed convincingly by entering the question into an AI tool and submitting the output, better detection may only ever treat the symptom.

Changing what we ask students to demonstrate tackles the underlying problem.

Security should then follow the purpose of the assessment. Once institutions are clear about what an assessment is testing, decisions about security become easier.

An exam testing unaided knowledge will need different safeguards from one in which selected digital resources are intentionally available. Security can then be proportionate to the risk rather than applied indiscriminately.

That may involve combinations of browser controls, appropriate proctoring, audit trails and human review. Crucially, technology should provide evidence to support academic judgement rather than attempting to replace it.

This matters because false suspicion has consequences. Institutions need reliable evidence when misconduct genuinely requires investigation, while students need confidence that legitimate behaviour will not automatically be treated as suspicious.

AI therefore presents education with a choice.

We can try to preserve every existing form of assessment and build progressively higher walls around it. Or we can use this moment to reconsider what assessment is supposed to tell us. The Bett findings suggest AI is becoming an established part of teaching practice. Keeping it entirely outside learning will become increasingly difficult.

Assessment needs to evolve accordingly.

For colleges and other education providers, the starting point should be to review assessments against one simple question: what does this assessment need a student to prove?

If it is testing foundational knowledge that a learner must possess independently, then AI should be excluded and the assessment environment secured accordingly. If it is testing analysis, judgement or the ability to apply knowledge, educators should consider whether controlled access to AI or other resources can be accommodated without undermining what is being measured.

From there, security should follow the level of risk. That could mean browser controls or proctoring for some assessments, while others may place greater emphasis on audit trails, explaining reasoning and human review. The important point is that institutions should not default to the same level of monitoring for every exam simply because AI exists.

This is not an argument for abandoning closed-book exams or allowing AI everywhere. It is an argument for being much more deliberate about when students must demonstrate knowledge independently, when using digital tools is itself a legitimate skill, and how we establish the difference.

AI is already changing how people learn and work. Education now needs assessments designed for that reality, not simply more technology for watching students take them.

By Dr. James Gupta, CEO of Synap