Some students use it to generate ideas. Others use it to challenge their arguments or improve their writing. And undoubtedly, some delegate almost the entire assignment to a chatbot.

What concerns me is no longer whether AI has entered the classroom; it has been there for a while. The more important question is whether we, as university lecturers, have truly learned how to guide students in such a learning environment.

This is not merely a personal observation.

A study published in 2026 found that 89.3% of Vietnamese university students surveyed reported using AI tools, yet only around a quarter had any formal training in AI literacy.

A global survey of 45,398 students and educators in 35 countries revealed a remarkably similar pattern.

While 88% of students reported using AI in their studies, 43% said they had never experienced AI being meaningfully integrated into any of their courses, and only 29% believed their lecturers were adequately prepared to guide students in using the technology.

In other words, students entered the age of AI long before universities had agreed on how to lead them through it.

That does not mean students understand AI better than their lecturers, nor does it suggest that educators are resistant to technological change.

The same global survey found that 77% of lecturers now use AI in their teaching, representing a 16-percentage-point increase from 2025.

Both students and lecturers are adapting.

The problem is that they are adapting in fundamentally different ways, while the shared rules governing AI use have struggled to keep pace.

Imagine a student taking five courses in a single semester.

The first course prohibits AI entirely, the second allows it but never clearly defines the limits, the third requires students to disclose any AI use, the fourth encourages AI use for brainstorming, and the fifth says nothing about it at all.

In this situation, students are not simply taking five different subjects; they are expected to navigate five different, and often unspoken, sets of rules.

The variation between each course is predictable. An academic writing course may have stricter limits on AI than a programming course. An examination designed to assess foundational knowledge cannot follow the same rules as a project intended to simulate a professional workplace.

The problem arises when the differences are driven not by systemic choices but by individual lecturers independently deciding how comfortable they are with AI and how much they are willing to tolerate it.

Thus, the same behavior could be considered good academic practice in one class and misconduct in another.

A student who uses AI to suggest the structure of an essay might be encouraged in course A, prohibited in course B, and left uncertain about whether the practice even needs to be disclosed in course C.

It becomes difficult for students to develop responsible habits of AI use when the learning environment itself provides no consistent language for distinguishing what is acceptable, what must be disclosed, and what students are expected to complete independently.

That, to me, is the most worrying guidance gap.

Banning AI altogether may seem like the simplest solution, but it is unlikely to solve the underlying problem. Students can still use AI outside the classroom, and the practice simply becomes less visible rather than disappear.

On the other hand, allowing AI without clear expectations is equally insufficient.

A student who knows how to open a chatbot and write an effective prompt does not necessarily know how to verify its answers, recognize fabricated references, or distinguish between an argument that merely sounds convincing and one that is genuinely supported by evidence.

However, the widespread availability of AI does not automatically translate into the maturity needed to use it responsibly.

But students themselves are evidently cognizant of this.

According to the Digital Education Council, 66% of students worry that AI could encourage superficial learning while weakening critical thinking and creativity.

Among lecturers, 73% fear students' reliance on AI could undermine the development of their skills.

Students are moving quickly, but are not always certain it is in the right direction.

Thus, a more useful question than "Should students be allowed to use AI?" may instead be "To what extent should AI be used in this particular assignment?"

Three college students talk on campus. Photo by George Pak via Pexels

Some universities have already begun answering that question in various ways.

Some have published policies on academic integrity and AI-assisted work, issued guidances on students' use of generative AI, or revised their academic integrity frameworks to require lecturers to specify the expectations and limits of AI use in major assessments.

Another serious initiative is the trial use of AI in assessment, where assignments follow a five-level framework ranging from no AI use to unrestricted AI use.

The value of this approach does not lie in the fact that there are five levels or its implication that every university should adopt the same model.

Its real value lies in changing the conversation.

Instead of debating whether AI should simply be banned or permitted, lecturers are encouraged to answer a different set of questions: What knowledge or skills is this assignment actually assessing? At which stages may AI be used? Which parts must students complete independently? How should AI use be disclosed?

An assignment designed to develop foundational writing skills may require students to write entirely on their own.

Another may permit AI to generate ideas or suggest revisions, provided students critically evaluate and improve the output themselves.

For a project intended to simulate a professional workplace, AI might even become an expected part of the workflow, as long as students can justify how they used it and accept responsibility for the final product.

The differences between courses remain, but under this framework they no longer feel like arbitrary rules that students must guess; instead, they become deliberate choices aligned with each course's learning objectives.

I believe this represents an important shift.

Universities do not necessarily need a single, uniform policy governing AI across every discipline. Different educational goals naturally require different approaches to AI.

What they need instead is a shared language, one that enables lecturers and students alike to understand different levels of AI use, disclosure expectations, and the boundary between technological assistance and the replacement of human thinking.

Nevertheless, even a shared language is not enough if it exists only on paper.

From my own experience in the classroom, I believe there are at least three practical steps that every lecturer, department, or even university can begin taking immediately and without waiting for comprehensive policies from the top.

First, lecturers should make expectations explicit from the outset instead of leaving them unspoken.

In the age of AI, a well-designed assignment should do more than describe the task itself, also specifying exactly where AI may and may not be used.

Can students use it to search for sources? To generate an outline? To check grammar? Or is AI prohibited entirely?

Assignments should explain how AI use should be disclosed.

Students might, for example, attach their conversation with a chatbot or include a brief statement describing which tools they used and for what purpose.

This level of clarity requires little more effort than writing a vague assignment brief, but eliminates much of the uncertainty that forces students to guess the rules.

Second, schools should redesign certain forms of assessment rather than simply tightening existing ones.

Many traditional take-home essays, completed independently and submitted online without supervision, can now be produced almost entirely by AI, regardless of whether it is allowed.

Instead of relying solely on stricter policing, universities should reconsider what they assess and how.

Some components of assessment could shift toward formats that are much harder to outsource to AI: a short oral defense of a submitted essay, an in-class writing exercise completed under time constraints, or a reflective account explaining the student's reasoning process rather than merely presenting the finished product.

The purpose is not to punish students for using AI.

It is to ensure that assessments genuinely evaluate the knowledge, judgment, and skills they are intended to.

Third, universities should teach students how to evaluate AI and not merely how to use it.

Most students today learn independently how to write prompts for chatbots.

Few receive systematic instruction on how to identify answers that sound convincing but are factually wrong, verify AI-generated information against reliable sources, or recognize when a chatbot has fabricated a citation that does not exist.

These are not specialized AI skills, just forms of critical thinking that can be incorporated into many different disciplines without requiring a separate AI course.

None of these three measures can replace the need for university-wide or national policies. But they do demonstrate that the guidance gap is not entirely beyond the reach of individual lecturers. Part of that gap can already be narrowed through the design of individual assignments and classroom practices, without waiting for a comprehensive regulatory framework.

It would be inaccurate to suggest that Vietnamese universities are standing still. The examples above show that some institutions have already introduced policies, issued guidance, or trialed new approaches to assessment.

At the same time, it is unrealistic to expect a handful of isolated initiatives to evolve into a coherent system of guidance on their own.

As AI advances faster than curriculum review cycles, faster than the development of institutional policies, and sometimes even faster than educators can learn to adapt themselves, individual effort remains necessary but is no longer sufficient.

The challenge runs deeper.

AI is not simply changing how students complete assignments; it is forcing universities to reconsider the very assignments they ask students to complete.

In one global survey, only 28% of students believed that most or many of their assessments reflected the work, skills, and judgment required in AI-enabled workplaces. The rest did not believe their assessments consistently aligned with those demands.

This raises a far more difficult question than whether students use AI to complete their coursework: If graduates will spend their professional lives working alongside AI, how should universities assess them today?

The answer is certainly not to incorporate AI into every assignment. Some capabilities can only be developed through direct intellectual effort—reading, writing, calculating, debating, and wrestling with complex problems independently.

But it is equally unjustifiable to design every learning activity as though the technology students will encounter every day after graduation simply does not exist.

The boundary between appropriate AI assistance and independent thinking must be intentionally designed; it cannot be left for students to figure out on their own.

Perhaps, then, the most important question is no longer how universities can keep pace with AI since it is a race they are unlikely to win.

What students need instead is an education system capable of showing them when AI genuinely extends human capability, when convenience begins to replace genuine thinking, which tasks can responsibly be delegated to technology, and which require human judgment, verification, and accountability.

The gap at the front of today's classroom is therefore not a shortage of new technology but a gap in guidance since students have already entered a new era of learning while the education system is still trying to determine how best to lead them through it.