The challenge: assessing knowledge and skills with AI out of reach
Designing assessment in the age of AI
For Irene Blankers, an educational advisor responsible for assessment at the Department of Industrial Engineering and Innovation Sciences, AI is not simply something universities should block. Students will use these tools throughout their future careers, and the department wants them to learn how to use them well.
The challenge is deciding when AI should be part of the learning process and when students first need to demonstrate that they have mastered the underlying knowledge and skills themselves.
"I guess that's the most difficult task in my job right now with AI," Irene says. "You want to make students AI literate, but you also need them to demonstrate that they have the basic knowledge without it."
"You want to make students AI literate, but you also need them to demonstrate that they have the basic knowledge without it.”
Irene Blankers — Educational Advisor at Eindhoven University of Technology
That distinction has become particularly important for take-home assessment. When every student has unrestricted access to AI tools such as ChatGPT and Claude, an assignment can still show what a student is able to produce, but it becomes much harder to establish what they can do independently.
The same problem applies to practical skills. A programming assignment completed at home may show that working code was produced, but not necessarily that the student can write that code without AI assistance.
"So if a course requires students to learn how to program or use a certain application, then I need a controlled environment to test that skill," Irene says.
Securing exams with applications
Before Schoolyear, the department used a USB-based solution called Digi-Step to secure its digital exams. Students brought their own laptops to campus, inserted a USB stick and entered a controlled environment where they could not access resources such as Google or AI.
In practice, the setup brought technical friction. Irene recalls recurring problems with the USB stick across different types of student laptops, particularly Apple devices. At one point, the compatibility issues became serious enough for the department to stop using Digi-Step altogether for a period of time. Some of those issues were later resolved, allowing the system to be introduced again, but the experience remained difficult.
There was also a more fundamental limitation. The setup worked for exams that did not require applications such as RStudio. But, like most exam security solutions, it could not give students access to those applications while also closing off the cheating routes within them, such as AI assistants and access to personal files.
Giving students access to a desktop application introduces a different integrity challenge. Applications can provide their own routes to AI assistants, personal files and cloud storage. Simply allowing the software to open does not necessarily create a secure application-based exam: the application becomes available, but resources inside or accessible through it may become available as well.
That created a gap for courses where the learning objective was not simply knowing the theory behind programming, statistics or data analysis, but being able to perform the work in the software itself.
The solution: secure exams with the applications students actually use
Schoolyear gave the department a way to bring those two requirements together.
Inside Schoolyear's Safe Exam Workspace, students can work in coding environments and applications such as Python and RStudio while potential cheating routes are blocked. Personal files, unauthorized websites, AI tools and other applications stay out of reach.
"What has been really nice about using Schoolyear is that students can use an application during an exam and we can make it AI-proof," Irene says. "This way, we can assess if they have the knowledge themselves."
"What has been really nice about using Schoolyear is that students can use an application during an exam and we can make it AI-proof. This way, we can assess if they have the knowledge themselves."
Irene Blankers — Educational Advisor at Eindhoven University of Technology
The department started the implementation with a Python course two years ago. The pilot gave the team an opportunity to test whether an application-based exam, in this example Visual Studio Code with Python could run securely on students' own laptops without changing the underlying assessment platform.
It went well enough for the department to expand the approach. Within a year, Irene redesigned the courses that had previously relied on Digi-Step and moved them to Schoolyear.
Rather than having to choose between a secured digital exam and an exam that gives students access to the applications they actually use, educators could now combine the two.
Keeping the existing assessment workflow in place
Moving to Schoolyear did not require educators to move their exams to a new assessment platform.
The department already used the assessment platform Ans, and teachers continued creating their questions and assessments there. Schoolyear integrates with any LMS or assessment platform like Ans and provides the secure environment around each exam, allowing the existing assessment workflow to remain largely intact.
For educators, that means the security layer can change without requiring them to rebuild the assessment process around another system.
Application-based exams with Schoolyear offer a high degree of flexibility in how the exam environment is configured, which means they require some preparation beforehand. The university's IT team works with educators to determine which applications, files, websites and other resources students need for the assessment. The secure environment can then be configured around those requirements.
Once that setup has been completed, both the educator's and the student's experience remain straightforward.
The student experience
Michelle Verrijth handles much of the operational setup of Schoolyear within the department together with her colleagues.
"Students take exams on their own devices, except if there are technical difficulties on exam day. In that case, they can use one of our loan laptops," Michelle says.
Students can complete a device check from home before the exam, allowing compatibility issues to be identified before they enter the assessment room. If a device cannot be used, the university maintains loan laptops as a fallback.
Technical issues can still occur, as they can in any BYOD environment, but they are now limited to a small number of students. Irene describes the difference with Digi-Step as significant: the device problems that had previously made the USB-based setup difficult to operate are no longer a recurring feature of each exam.
During the assessment itself, students only get access to the resources the educator has decided they need.
Depending on the exam, students may get access to attachments, websites or applications. "The only things they can open are these whitelisted resources," Irene says. “Everything else is blocked.”
A student might receive a programming problem in Ans, use Python and VS Code to solve it and submit the resulting code. In another course, the student might receive a dataset, perform the analysis in RStudio and submit the output at the end of the exam.
The result is a controlled exam environment without stripping away the tools students need to perform the actual task.
The result: assessment that reliably measures the skill being taught
Giving students access to applications inside a secure environment changes what teachers can ask them to do.
Instead of limiting an exam to multiple-choice questions, short written responses or questions about how students would approach a task, educators can ask students to carry out the task itself. Students can write code or perform an analysis and submit the output they have produced.
That makes the assessment much more authentic. If a course is designed to teach students how to work with Python or RStudio, the exam can evaluate that ability directly.
A clearer picture of what students know without AI
The Safe Exam Workspace has made differences between supervised and unsupervised assessment more visible.
Irene sees this in courses that combine take-home work or unsecured midterms with a secured final exam. Students may perform well when they have unrestricted access to outside resources such as AI, but the final exam creates a point at which they have to demonstrate the same underlying knowledge independently.
"With midterm exams, a number of students score higher grades because these are not yet AI-proof,” she says. "But when the final exam comes, we see a lower passing rate among these students."
"With midterm exams, a number of students score higher grades because these are not yet AI-proof. But when the final exam comes, we see a lower passing rate among these students."
Irene Blankers — Educational Advisor at Eindhoven University of Technology
For the department, that difference is useful beyond academic integrity alone. If students appear to understand the material during less controlled coursework but struggle once they have to perform independently, the result can help expose where the course itself may need to change.
The secured assessment therefore does more than establish who can complete the task without outside assistance. It also gives the department a clearer signal of where students may need more teaching, practice or preparation before they have truly mastered the skill.
What's next: verifying skills earlier in the learning journey
Until now, the department has primarily used Schoolyear for final exams. It is now testing the approach in three midterm exams within one course.
Behind that experiment sits a broader question about the role of assessment in the age of AI: at what point should students be required to demonstrate a skill independently, and when can AI become a legitimate part of the way they apply it?
Irene's answer is not to remove AI from education.
Instead, she sees value in creating deliberate points in the learning journey where students first demonstrate that they have acquired the underlying knowledge or skill. Once that foundation has been established, AI can play a different role: not as a substitute for knowledge, but as a tool students use on top of knowledge they have already demonstrated.
"If students need to be able to program, then at the beginning of the learning journey I need to make sure that the students know how to do it without help from AI," Irene says. “And after that, it's OK if they use AI because they've already showed to me once that they can do it themselves.”
.jpg)
