Better Evidence, Not Better Detection
Reflections on the latest Wonkhe and Studiosity report on AI, learning and academic integrity.
Wonkhe and Studiosity’s new report, Keeping the learning real, is one of the clearest accounts yet of where the sector stands on AI and academic integrity. It is also candid about where the sector still falls short.
The headline is encouraging. Institutions have moved on from prohibition and detection. They now accept AI as part of study, invest in AI literacy and redesign assessment from first principles. Progress has been real and fast.
But whether AI is permitted or prohibited, the fundamental responsibility remains the same. Institutions create opportunities to learn. Students create evidence of that learning. Our job is to ensure that evidence is authentic, credible and sufficient to show that meaningful learning has taken place.
But the report is equally clear about what remains unsolved. How do institutions know learning has actually happened?
Vivas are robust but rarely scale. Declarations become mechanistic and can feel like policing when guidance is unclear. The authors place the residual risk squarely here.
We think they are right. We also think the implications go well beyond academic integrity.
The question has changed
The first phase of the AI debate asked whether students had used AI. That question was always going to have a short shelf life. Detection is unreliable and the cost of getting it wrong is high. YouGov polling for Studiosity found three-quarters of students anxious about being wrongly accused of plagiarism over their use of AI. The risk is not simply student anxiety. It is a system that encourages learners to prove they have not cheated rather than demonstrate what they have learned. When confidence depends on suspicion, trust erodes. When confidence is built on evidence, learning becomes visible and authenticity becomes easier to establish.
The more durable question is whether there is enough evidence to be confident that meaningful learning has taken place. That question sits at the heart of assessment, regardless of how technology evolves. Institutions create opportunities to learn. Students create evidence through the process of learning. The challenge is ensuring that evidence is authentic, credible and sufficient to support confidence in the outcome.
That is a different problem. It cannot be solved at the point of submission. A finished essay shows what a learner produced. It rarely shows how understanding developed, how feedback was applied or how judgement was exercised along the way. That matters because authenticity does not live in the final artefact alone. In an AI-enabled world, confidence comes from understanding the process behind the outcome. The more visible the learning journey, the easier it becomes to establish that meaningful learning has taken place.
Confidence comes from connected evidence
The report’s central recommendation is integration. AI literacy, open and closed assessment and learning validation should reinforce one another, held together by a culture of trust.
We agree. And we would add a second thread: evidence. Trust is easier to build when learning can be proven and authenticity can be established with confidence.
Confidence rarely comes from a single checkpoint, declaration or viva. It comes from the process of learning itself. The development and feedback cycle, where learners practise, receive feedback, reflect, improve and apply their knowledge, creates evidence over time. When that evidence is connected across modules, programmes and experiences, authenticity becomes easier to establish and learning can be proven with confidence.
This is not an argument for surveillance. Institutions do not need more data. They need better evidence that supports better judgement, the case we made in Assuring Learning in the Age of AI and Assured Assessment. Evidence gathered as part of learning feels like learning, not policing.
One body of evidence, three kinds of confidence
What makes this shift significant is that the same evidence serves more than one purpose.
Evidence of learning helps educators understand how a learner has developed and make defensible judgements about outcomes.
Evidence of readiness helps learners show what they know and can do. As we argued in World Readiness Requires Evidence, learners cannot communicate development they cannot see. A qualification still matters, but employers increasingly want to see what sits behind it. That theme ran through our response to the Future of the American University report.
Evidence of quality helps institutions demonstrate assurance of learning to accreditors and regulators. It also shows whether programmes are achieving what they claim. The evidence that explains one learner’s progress can explain whether a programme is working.
These have long been treated as separate agendas. Different teams own them. Different systems support them. Yet the same evidence sits behind learning, readiness and quality. AI is forcing these traditionally separate conversations together. Assessment, integrity, employability and quality assurance are converging on one shared need: evidence that is connected, credible and meaningful to everyone who relies on it.
What comes next
The sector has done the hard early work. It has moved beyond bans, built literacy and redesigned assessment at pace. The next phase must be less about controlling the tools students use. It is about how institutions capture, connect and use the evidence learning creates, not only to establish authenticity, but to support employability, assure quality and build confidence in learning itself.
That will require new thinking about where evidence lives, who it belongs to and how it moves across an institution and beyond it.
The defining question will not be whether a student used AI. It will be whether there is evidence that learning took place.
The answer is not better detection. It is better evidence.
