There’s a phrase I’ve returned to often in conversations this year: “What are we actually assessing?” It’s a deceptively simple question—one that predates generative AI, but one which now demands renewed urgency. Over recent weeks, I’ve been engaged in a challenging and generative exchange with a colleague about the implications of AI for assessment integrity. The conversation was thoughtful, well-informed, and at times starkly divergent in its conclusions. What emerged was not just a technical debate about proctoring or misconduct, but a deeper contest of educational values.

My colleague argues, with growing conviction, that remote assessment is already untenable. AI systems like GPT-4 can simulate human output with such fluency and consistency that, from their perspective, no student work submitted remotely can be considered trustworthy unless some element of in-person interaction is involved. Their proposed solution is clear: mandate in-person components for every assessment, even if those components are brief, such as face-to-face supervision meetings or viva-style check-ins. Without such mechanisms, they argue, our degrees will lose their value, our integrity will falter, and our institutional reputation will suffer.

There is something deeply principled in this view—an unflinching commitment to credibility and fairness. But there is also, I would argue, a narrowing of the university’s educational horizon. If our response to AI is to reframe assessment as a process of authentication, we risk mistaking verification for learning. We retreat from the relational, dialogic, and transformative dimensions of higher education into a posture of defence.

What this perspective highlights most clearly is the fragility of many of our current assessment designs—how easily they can be mimicked or subverted when reduced to outputs. But that fragility predates AI. If a chatbot can complete an assignment convincingly, we must ask whether the assignment truly invites critical thinking, synthesis, or personal meaning-making. Our challenge is not to make assessments AI-proof, but to make them worth doing.

This is not a call to ignore the risks. Generative AI does pose serious threats to academic integrity, and we would be naïve to assume goodwill alone will suffice. But nor should we re-engineer assessment policy around the assumption that students are seeking to evade learning at every opportunity. Education cannot flourish under surveillance. Nor can it rely solely on physical co-presence. Accessibility, flexibility, and care are not add-ons to rigour—they are part of how we define it.

So what might a more constructive response look like?

First, we need to re-centre assessment within learning. This means moving away from static tasks and towards dialogic, contextualised processes—reflective portfolios, community-based projects, scaffolded submissions with feedback loops, and tasks embedded in students’ lived experiences. These forms of assessment are not invulnerable to AI, but they are more resistant to substitution because they are anchored in relationship and development over time.

Second, we need to invest in staff support. This is not simply a matter of academic integrity protocols—it’s a matter of pedagogical design. Colleagues need time, guidance, and institutional recognition to rethink what assessment could and should be in an AI-saturated world.

Finally, we need to act collectively. Assessment integrity cannot be maintained by isolated academics working in private unease. Nor can it be outsourced to commercial proctoring systems promising “robustness” through opacity. We must cultivate spaces of critical dialogue across disciplines, roles, and student communities to shape our shared response.

AI is not just a threat. It is a mirror. It reflects the assumptions embedded in our assessment systems—some of which are long overdue for scrutiny. If our goal is to preserve what is valuable in higher education, we must begin not by locking the doors tighter, but by opening the conversation wider.