Teaching Ethics: One Year Later, Why Are More Students Using AI When They’re Told Not To?

A year ago, I wrote about a teaching activity that explored whether helping students think more explicitly about ethical behaviour might influence the way they used Generative AI (GenAI) in assessment. The activity was designed as an assessment for learning experience, as part of normal teaching practice. Students had engaged with professional ethics and responsible AI use, and they were then asked to complete a personal reflective assessment in which AI use was not permitted (even though I knew many would anyway). Afterwards, they were given an opportunity to reflect on how they had approached the task and what they might do differently in future.

The process is outlined in detail in the previous post, but to help anyone interested in reproducing the learning activity, a visual representation may help:

In 2025, when I reflected on the responses, I came away with several insights about student behaviour, ethical preparation, and the growing role of AI in assessment. The 2026 responses have made me revisit some of those conclusions. In some cases, they reinforced what I thought I was seeing. In others, they made me rethink my assumptions and reconsider what might actually be driving student behaviour.

Students seem to understand the expectations

One of the first things that stood out was that fewer students said they were unaware that the tools they had used were not permitted. That dropped from 13% in 2025 to 6% in 2026.

AI used for reflection	2025	2026
Stated they had not used AI	9%	9%
Stated being unaware that the tools they had used were not permitted	13%	6%
Openly admitted to using AI	78%	85%

At the same time, the proportion openly admitting to using AI increased from 78% to 85%. That combination is hard to ignore. If students were simply confused about what was allowed, we might expect clearer expectations to reduce inappropriate use. Instead, what seems to be happening is almost the opposite: awareness appears to be improving, while AI use is increasing. This corresponds with a recent UK report where 94% of students said that they use generative AI to help with assessed work, although, that figure includes assessments where AI may have been permitted or encouraged.

It is clear the problem is not simply that students do not understand the rules. It is becoming harder to argue that misunderstanding is the main explanation. Most students did not report being unaware of the restriction, yet AI use remained high. This reinforces my 2025 thoughts that clearer wording, stronger warnings, or more explicit instructions are unlikely to be enough on their own. As I have previously outlined, unsupervised assessment can still have considerable value for learning, but I no longer think we can rely on it alone when we need to assure independent capability. In an unsupervised environment, we cannot confidently know who or what produced the work.

Ethical preparation does not seem to be the problem

The responses around ethical preparation were almost unchanged.

In 2025, 100% of students indicated that they felt ethically prepared to use AI. In 2026, that figure was 99%. So again, at least from the students’ perspective, lack of ethical preparation does not seem to be the problem. This also made me realise that knowing a rule and accepting the ethical rationale behind that rule are two different things. A student may understand perfectly well that AI has been prohibited, while still questioning whether using it is actually unethical.

I still think teaching ethics is important. Actually, more important than ever before! Students need to think about accountability, authorship, transparency, professional responsibility, and what happens to their learning when they rely too heavily on AI-generated content. But the responses reminded me of something fairly obvious: understanding what the ethical choice might be does not automatically mean someone will make that choice.

There are always other factors involved. Ethical decision-making also does not happen in a vacuum. Students operate in environments where convenience, competition, incentives and self-interest all shape behaviour, and where the conduct they observe in professional and public life does not always match the ethical ideals we teach in the classroom. Furthermore, convenience matters. Time matters. Perceived quality matters. Habit matters. Personal judgement matters.

So perhaps the better question is not whether ethical preparation works. Perhaps it is whether we have been asking ethical preparation to do too much.

Why are students using AI?

The reasons students gave for using AI are where the 2026 responses started to challenge my thinking most.

Motivation to use AI (selected responses & can select more than one option)	2025	2026
Believe that AI would improve the quality of my work	46%	44%
Lacked confidence in my ability to complete the task without AI	15%	5%
Used to save time	14%	18%
I believe it's a legitimate learning tool	13%	16%

The most common motivation hardly changed. In 2025, 46% said they believed AI would improve the quality of their work. In 2026, that figure was 44%. These figures mirror a trend in a much larger 2026 UK student survey that found that 47% used AI for this reason (down from 50% in 2025).

I am actually surprised that this figure is not higher. Many students are in competitive mode and do want to maximise every mark they can get; they learn this in their education trying to get into university. If used correctly, it can definitely improve the final product. But the used correctly part is a big part of the problem, and I see that in student AI use all the time! They need guidance and awareness. The two percentages are so close that I would not read much into the decline itself. What interests me more is that fewer than half selected quality improvement at all.

Further to my surprise, what changed much more was confidence. The proportion of students saying they lacked confidence in their ability to complete the task without AI fell from 15% to just 5%. Looking for possible reasons, I found this recently published pre-print that showed how AI advice affects people’s willingness to admit uncertainty. Across five experiments, access to AI advice dramatically reduced ‘I don’t know’ responses, even when the AI advice was deliberately wrong. Participants answered more questions, were less accurate, and reported substantially greater confidence.

Using AI to save time increased from 14% to 18%, while the proportion describing AI as a legitimate learning tool rose from 13% to 16%. Those are relatively small changes, so I would not want to build too much around them on their own. But again, I am surprised by the 'saved time' component being so low. The same 2026 UK student survey found that saving time was recognised highly, but had actually trended down from 51% to 45% of students. The questions, cohorts and contexts are different, so I wouldn't directly compare the percentages, but the contrast still made me wonder why saving time featured so little in these responses.

One possibility is that the mindset is shifting from "I cannot do this without AI" towards something closer to: “I could do this without AI, but why would I?”

That is quite a different form of reliance. Someone can feel perfectly confident in their own ability and still become accustomed to outsourcing parts of the task. Over time, the interesting question may not be whether students think they are dependent on AI, but whether their normal ways of working are becoming dependent on it. Of course, cognitive offloading is not inherently problematic. We routinely use tools to reduce cognitive effort. The educational question is whether the tool is removing incidental effort or replacing the thinking we actually want students to practise.

That feels much harder to see, and probably much harder to address.

What happens after students reflect on their behaviour?

The aim of the designed assessment activity was never to catch students using AI. It was to get them to think about the choice they had made, why they made it, and what that might mean for their learning and future professional practice.

Implication of the assessment on practice	2025	2026
On reflection, maintained that AI is a valid tool and should be embraced, not avoided	43%	45%
On reflection, stated that this assessment for learning process motivated them to do better & be more thoughtful in the future	48%	42%

In 2025, 48% said the process motivated them to do better and be more thoughtful in the future. In 2026, that figure was 42%. At the same time, the proportion who maintained that AI was a valid tool that should be embraced rather than avoided remained almost unchanged: 43% in 2025 and 45% in 2026.

Initially, it might be tempting to see those two responses as opposites. I am not sure they are. A student can become more thoughtful about their AI use while still believing AI is a legitimate learning tool.

The objective should not necessarily be to convince students that AI is bad. The objective should be to help students think more carefully about when it helps, when it harms, and when its use changes what they are actually learning.

That is a much harder conversation than “AI allowed” versus “AI not allowed”. The reality that we can't put the genie back in the bottle is becoming clearer every day. The broader evidence points strongly towards increasing and increasingly normalised student use of AI. That makes repeated opportunities for students to examine both the benefits and the risks increasingly important, not as warnings about AI, but as practice in judging when AI is supporting their thinking and when it is replacing it.

A way forward

This is why I developed and have been sharing the CAC Co-intelligence framework. If facilitated correctly, it helps students use AI in a supportive, augmented approach, helping them understand their own learning and placing themselves in the driver's seat of their own thinking. Most importantly, it encourages them to take responsibility for their learning/outcomes and remain in control.

Perhaps students are right to challenge some of our boundaries. If AI becomes a legitimate professional tool, we need good reasons for asking students not to use it. “Because the assessment says so” is unlikely to remain a convincing educational argument. Sometimes there is a very good reason. We need to know that a student can perform a particular task, make a judgement, or demonstrate a capability independently. In those situations, we need assessment designs that give us credible evidence of that capability.

At other times, perhaps the better question is not whether AI was used, but how it was used, what thinking remained with the student, and whether its use enhanced or diminished the learning we were trying to achieve. That is why I think engagement matters. If AI use becomes habitual before students have developed the judgement to recognise what should be offloaded and what should remain theirs, changing those habits later may be much harder.

Perhaps the challenge is no longer teaching students simply when they can and cannot use AI.

It is helping them develop the judgement to know when they should.

Sasha Nikolic

28/07/2026