According to the study Use and Perception of AI in the University Environment by Fundación CYD (2025), 89% of Spanish undergraduate university students use some form of generative AI, and 35% use it on a daily basis.
In other words, in most institutions, AI is already part of students’ work, whether the institution has decided how to address it or not.
And although this data comes from higher education, anyone who has spent time in a vocational education classroom knows that the situation is very similar.
The problem is the absence of institutional criteria for this use.
Without shared criteria, individual teachers make decisions that can contradict one another: one bans it, another allows it under certain conditions, another ignores it…
Students then do not learn to develop their own judgement. They learn to navigate ambiguity, which is a very different—and far less useful—skill.
Why academic integrity policies need to be updated
Most institutions have academic integrity policies that were written before generative AI was accessible to students.
These policies address plagiarism, cheating and source citation, but they do not account for a scenario in which a tool can generate text, code, outlines or even solutions to problems within seconds.
Updating an academic integrity policy to address AI use does not mean banning the technology. It simply means establishing the framework within which learning takes place and defining when and how AI use makes pedagogical and ethical sense.
It is important to rely on solid references. Article 4 of the AI Act establishes the obligation to take measures to promote AI literacy among staff.
Three open-access frameworks also provide guidance on the competencies an educational institution should develop: UNESCO’s AI Competency Framework for Teachers and Students (2024) and the AILit Framework developed by the OECD and the European Commission (2026).
These are not legally binding standards, but they are among the strongest reference points available for developing a policy based on evidence rather than building one blindly.
Which uses are acceptable and in what context
Not all uses of AI in academic work are equivalent.
Using AI to search for information is different from using it to write the argument. Using it to generate code is different from using it to understand code you have already written. Using it as a spellchecker is different from using it to produce the entire assignment. That is why an effective policy should not simply say that AI is allowed or prohibited. It should define which types of use are acceptable for which types of task and who has the authority to establish exceptions within a particular subject.
A binary rule may reassure the person who writes it, but it does not work for the person who has to apply it in a real classroom.
What does the institution ask students to do when they use AI?
Some institutions are introducing AI-use declarations as part of assignment submissions: students indicate whether they have used AI, for which part of the assignment and in what way.
This approach has a pedagogical advantage that goes beyond monitoring. It requires students to become aware of their own process, not just their final result. And that awareness is, ultimately, the first step towards developing the judgement we want students to build.
Proportionate consequences, not just punishment
Consequences for unauthorised AI use should be proportionate and educational, not purely punitive. A student who submits an assignment generated entirely by AI without declaring it has done something different from a student who uses AI to review their final writing. Treating both cases in exactly the same way is pedagogically counterproductive.
And there is something worth saying plainly. Banning AI without a clear pedagogical rationale does not protect academic integrity. It simply pushes AI use into a place where nobody can see it.
The problem with a detection-first approach
When institutions feel pressure to do something, many turn to AI-content detection tools. These tools have serious limitations: significant false-positive rates, the ease with which their results can be circumvented through minor modifications, and no inherent pedagogical value. Accusing a student based on the result of a detector that can be wrong is a risk no educational institution should take.
Making detection the primary strategy shifts the focus from teaching to catching students out. And in a context where students have constant access to AI tools in their everyday lives, this is a race the educational institution cannot win.
The alternative is to design tasks that make the learning process visible, not just the final result: assignments completed in stages with intermediate reflection, oral presentations, process portfolios or real-time argumentation. This type of design makes AI less relevant as a shortcut because what is being assessed cannot simply be delegated to a machine.
How to build the policy step by step
- The first step is diagnosis: understand what is already happening in the institution. Not what should be happening, but what is actually happening. This means speaking to all educational stakeholders, including students, not just teaching staff, because the people who know best how AI is being used in assignments are the people using it.
- The second step is defining shared criteria: a working session with teaching staff to agree on the principles that will guide the policy before drafting any rules.
- The third step is drafting: a short, clear and practical document that can be read in five minutes and applied in the classroom. Not a twenty-page regulation, but a guide to shared criteria.
- The fourth step is communication: explain the policy to students and families, not as an imposed set of rules, but as a framework for digital coexistence built by the whole community.
What is really at stake
It is easy to experience all of this as a threat.
Students are cheating. Technology is getting ahead of us. We need to defend ourselves.
But I have spent enough years in classrooms to know that behind a student using AI without judgement, there is usually not bad intent. There is often simply a lack of a framework that nobody has given them.
Academic integrity is not protected by policing students. It is protected by teaching them how to make decisions. And a well-designed AI policy is not a set of rules for catching cheaters, perhaps it is one of the ways an institution tells its students that it trusts their ability to use technology responsibly—and that it will support them in developing that ability.
That is the difference between an institution that fears AI and one that integrates it with judgement.
Frequently Asked Questions
Can students use ChatGPT in their assignments?
It depends on what the educational institution has decided.
Without a clear policy, teachers make individual decisions that can contradict one another.
An academic integrity policy establishes shared criteria that provide clarity and security for students, teachers and families.
How can you detect whether a student has used AI in an assignment?
Automated AI detectors have significant false-positive rates and can be easily circumvented.
The better strategy is not detection, but designing tasks that make the learning process visible.
When an assignment requires real-time argumentation or a process portfolio, AI becomes much less useful as a shortcut.
What should an academic integrity policy on AI include?
Clear criteria defining permitted and prohibited uses by type of task, a process for students to disclose their use of AI, proportionate and educational consequences, and a framework for periodic review because the technology changes.
The policy you write today will need to be reviewed within twelve months.
Sources
- Fundación CYD (2025). Use and Perception of AI in the University Environment. Fundación Conocimiento y Desarrollo. Survey of 800 undergraduate students at Spanish universities.
- UNESCO (2024). AI Competency Framework for Teachers. United Nations Educational, Scientific and Cultural Organization. Open access.
- OECD and European Commission (2026). AILit Framework. AI literacy framework for education. Open access.
- Regulation (EU) 2024/1689 of the European Parliament and of the Council (Artificial Intelligence Act), Article 4, as currently applicable following Regulation (EU) 2026/1744.
