AI use is near-universal, but university policy lags behind

The takeaway: large classes and thin teacher contact make university adaptation harder, not easier.

Can detection solve academic integrity?

Teachers have used AI tools to check submissions for AI-generated text. Universities are now moving away from that approach because the technology is not always accurate.

Experts warn that relying on such detectors alone can lead to false accusations of student misconduct.

The debate has widened beyond cheating to student offloading: handing work to AI so completely that the student learns little or nothing.

Researchers and policymakers acknowledge that the scale of this behavior is difficult to gauge.

The July study shows what institutions published, but it cannot measure how often students offload work or whether staff apply policies consistently.

Scientific Editor Lisa Lock wrote: “Students cannot be forced to learn through a policy.”

The takeaway: detection cannot prove learning, and a rule cannot make a student learn.

What changes in assessment?

Universities have mainly responded by redesigning assessments rather than building broader system-wide support.

• Shift more assessment to in-person exams and oral presentations where AI cannot easily be accessed.
• Allow AI in some take-home essays, including cases where students cite their use of the technology.

Redesigned assessments can show whether learning has occurred, but they are not enough on their own. What students are assessed on also matters.

The takeaway: changing the task can expose learning, but the learning target still needs careful design.

How should students learn to use AI?

In the European Union, attention is shifting toward explicitly building students’ capacity to adapt. This includes teaching them to:

• Judge how much they know about a subject.
• Critically evaluate information with and without AI.
• Make decisions about learning processes and goals independently of AI or an educator.

These capabilities are hard to develop and mostly do not transfer between subjects. Institutions must teach them explicitly within the relevant discipline; specific course modules are one option.

In 2023, Lock and other academics wrote evidence-based assessment principles for the Australian higher education regulator.

The first principle requires assessment to show that students learned the content rather than offloaded the work to AI. Assessment redesign under this principle is already having an effect.

The second prepares students to work in a world where AI is ubiquitous. It has proved harder to implement because skills tied to today’s tools become outdated as the technology advances.

Simple “AI literacy,” meaning basic skill in using the tools, is not enough.

The takeaway: assessment reform must teach AI judgment within each discipline, not merely police tool use.

Can universities solve this alone?

There is no single approach that fits every university. Collaboration would help, but institutions tend to compete rather than share resources. Only one of the 39 mentioned learning from another institution.

In 2024, the higher education regulator required every university to submit AI action plans and later published strategy guides.

A regulator can set expectations, but it cannot fund the long-term work of changing policy and practice across the country.

The Office for Learning and Teaching funded teaching innovation and helped universities share what worked. It closed in 2016, and a promised replacement never materialized.

A national approach would require an enabling body, independent of any single university, to share evidence and guidance across the sector.

Ireland’s Higher Education Authority offers a possible model; it has published a national framework on AI in teaching and learning.

Setting up such a body would be difficult. Universities guard their autonomy, funding remains unresolved, and a national authority could struggle to keep pace with technology that changes faster than policy.

In Lock’s view, universities risk becoming redundant in an age of intelligent machines unless they engage deeply and collaboratively. The practical next step for institutional leaders is shared assessment design, explicit student instruction and stable funding for long-term change.