In 2026, the digitization of university entrance exams is no longer an exception: many selection processes combine in-person tests and “at-home” tests, often for reasons of scalability, cost, and speed of grading. For those who teach, this transition raises a central pedagogical and institutional question: how do we safeguard merit when the exam environment is no longer controlled by the university, but by the student’s domestic and technological ecosystem?
In this scenario, the expression comes into playoff campus ai: the use of artificial-intelligence-based assistants off campus (and outside institutional systems), accessible from personal devices, in real time. We’re not talking only about “studying with AI” (which can be legitimate), but about the possibility of receiving support during a selection test, in an undeclared and unverifiable way.
Why “at-home” entrance tests change the rules of merit
When a test moves remote, the very nature of comparability changes. In a classroom, conditions are relatively uniform: same constraints, same time, same supervision, same infrastructure. At home, instead, the environment becomes variable: connection quality, quiet, space, hardware, presence of other people, access to additional devices—and, above all—access to advanced digital tools.
This variability produces an asymmetry that directly affects meritocracy and access: those who have digital skills, familiarity with prompts and workflows, or simply a better setup, can gain an advantage unrelated to the disciplinary skills required by the program. In other words, the risk is that the test also measures (or above all) the ability to manage the technological environment and shortcuts, not real preparation.
For instructors and committees, the point is not to demonize technology, but to recognize that selection is a high-stakes act: it determines educational opportunities, social mobility, and often access to regulated professions. If conditions cannot be controlled, then test design must become more robust, and the rules more explicit.
Real risks: cheating, collusion, and new forms of academic integrity violations
The issue ofacademic integrity universitybecomes more complicated when the test is remote: opportunistic behaviors can be easier to carry out and harder to prove. It is useful to distinguish between improper use of AI during the test and legitimate use of AI in studying. The gray area arises when students do not have clear guidance on what is allowed and what is not, or when the test itself “invites” external support because it requires answers that are easily obtained via a search or a generative assistant.
The most frequent abuse scenarios in online selection processes (and incheating entrance tests) include:
- Second device: a smartphone or tablet out of frame to consult apps, browsers, chats, or AI assistants.
- External assistance: a person in the room or on a call who suggests answers or solves exercises.
- Real-time AI: generative tools used to produce answers, explanations, or calculation steps during the test (off campus ai).
- Shared databases: groups that collect recurring items, screenshots, solutions, and “strategies” to maximize the score.
- Asynchronous collusion: candidates who take the test in different time windows and pass content to those who have not started yet.
Alongside these risks, there is a fully legitimate use of AI: practicing with quizzes, clarifying concepts, receiving feedback on mistakes, building concept maps. The boundary line, from a pedagogical standpoint, is defined by two criteria:transparency(the student discloses tools and methods) andalignment with the objective of the test(measuring individual competencies, not the ability to delegate). If these criteria are not made explicit, the gray area becomes inevitable and the perception of unfairness grows.
Proctoring and AI detection: what works, what doesn’t, and which trade-offs to accept
The topic ofentrance exam proctoringis often presented as a “technical” solution to a complex problem. In reality, remote proctoring reduces some behaviors but incentivizes others (more sophisticated ones) and introduces non-trivial costs: surveillance stress, false alarms, accessibility barriers, privacy issues, and, at times, discrimination linked to housing conditions or neurodivergence.
In terms of effectiveness, proctoring is more reliable at detecting “physical” violations (moving away from the screen, presence of other people) than the use of discreet digital tools. Moreover, interpreting alerts requires trained staff and review procedures: without an audit process, there is a risk of turning a weak signal into an unjust sanction.
Evenai detection university selectiontools have structural limits: they work probabilistically and can producefalse positives(genuine text flagged as AI) andfalse negatives(generated text not detected). In an entrance exam, where errors have high consequences, basing decisions on a “likely AI” score is methodologically fragile and potentially challengeable.
A workable path is arisk-basedapproach, which calibrates control measures based on three variables: (1) the stakes of the selection, (2) the likelihood of abuse for that test format, (3) the impact on privacy and accessibility. In practice: the more decisive the test is and the more “easy to delegate” it is, the more robust measures are needed; but these measures must be proportionate, transparent, and accompanied by reasonable alternatives.
Examples of proportionate measures (which can be combined) include: shorter and synchronized exam windows, randomized items from a large bank, questions that require intermediate steps, spot-check oral verification, and clear procedures for contesting proctoring alerts. The key point is that technology does not replace design: it supports it.
Rethinking tests and grading criteria: designing AI-robust selection processes
To make a selection process more resistant to external assistance, it helps to shift the focus from “recognizable” answers to performances that demonstrate understanding, reasoning, and transfer. Pedagogical evidence on assessment suggests that authentic tasks, item variation, and process evaluation reduce the usefulness of copy-paste and increase the validity of the measure.
Concrete strategies, applicable even in standardized or semi-structured tests:
- Large item bank and continuous updating: reduces the circulation of “known” questions and rote memorization.
- Parametric randomization: same objectives, different numerical data or contexts (useful for math, logic, statistics, chemistry).
- Applied questions and “justified choice”: not only selecting an option, but indicating why with a brief, verifiable step.
- Targeted time constraints: tighter time on “delegable” items and more time on reasoning items (where thinking is needed, not searching).
- Reasoning trace: requiring intermediate steps, assumptions, plausibility checks, or an explanation of a common error.
- Targeted oral component: a short verification interview on 2–3 completed items, especially for anomalous scores or as a spot check (reduces collusion and delegation).
In particular, the short oral component can be designed as aentrance oral exam simulation(or post-test oral verification): not an “encyclopedic” interview, but a conversation focused on the process. Useful questions are: “Why did you rule out option B?”, “Which step would change if the data were different?”, “What is the most likely mistake here?”. These prompts are hard to delegate in real time and increase the validity of the score.
To make grading fairer, it helps to make explicit rubrics that value:coherence,justificationandcorrect use of concepts and procedures. Even in a multiple-choice test, you can include micro-spaces for explanation on a subset of items, without turning the test into an essay: a few well-designed characters can make the difference.
Responsible preparation: how to integrate StudierAI without penalizing those who use AI correctly
A frequent mistake in policies is treating AI as a single block: “banned” or “free.” In reality, to protect merit it is better to distinguish between preparation and performance. In preparation, tools likeStudierAIcan support studying with activities that, if well guided, increase autonomy and metacognition: controlled summaries, flashcards, adaptive quizzes, alternative explanations, error analysis, and alsoentrance oral exam simulationexercises to train presentation and reasoning under time constraints.
From a teaching standpoint, integrating AI responsibly means teaching students to: verify sources, compare solutions, justify choices, recognize hallucinations and limits, and turn a suggestion into learning (not delegation). This approach also reduces the likelihood that, in a selection setting, the student perceives AI as “necessary” to compete.
For committees and instructors, a workable policy can include three simple but very effective elements:
- Explicit rules for the test: what is allowed, what is forbidden, what is “not applicable” because the test is designed not to require it.
- Disclosure in preparation (if required): asking the student to declare how they used AI in exercises or a portfolio, without penalizing correct use.
- Consistency between preparation and test: if the test assesses reasoning, then preparation must also train reasoning (not just answers).
From an equity perspective, it can be useful to suggest accessible, guided study tools to students, instead of leaving everyone to fend for themselves with opaque solutions. If a class or department decides to indicate a common support tool for practice and self-assessment, the goal is not to “give an advantage,” but to reduce the asymmetry in digital skills and make preparation more transparent. Those who want to explore a training path cansign up for freeand set up study routines with quizzes and simulations, keeping it clear that different rules apply in the exam setting.
An operational summary for instructors: from total control to assessment robustness
With off campus ai, the realistic goal is not to reproduce at home the same control as in the classroom, but to increase the robustness of the selection process: design tests that measure authentic competencies, make rules and consequences transparent, and use proctoring and ai detection only as components of a broader system. In this way, merit is protected without imposing disproportionate costs on privacy and accessibility.
An essential checklist, useful for a collegial review of selection processes:
- Can the test be “delegated” to an AI assistant? If so, which parts can become more applied or process-oriented?
- Are there sufficient item banks and randomization to reduce the circulation of questions?
- Does proctoring (if used) have procedures for review, appeals, and handling false alarms?
- Is there a targeted oral verification planned, as a spot check or for anomalous cases, to strengthen the validity of the score?
- Are the rules on AI and permitted resources communicated clearly and with examples, before the test?
Finally, let’s remember that the most effective prevention often comes from instructional coherence: if throughout the year a critical use of AI for learning is promoted (not as a substitute), then the culture around the test changes too. If you want students to try guided study activities (quizzes, flashcards, oral simulations) transparently, you canstart for freeand, to understand the project’s approach and principles, consultabout us. The goal is not to chase the latest technology, but to preserve the meaning of merit in a context that has already changed.
