Reconsidering the value of exams in the age of AI

As the reality of Gen-AI settles across our campuses, many instructors have returned to in-person assessments. In a December 2025 Nature article, Vitomir Kovanović, Abhinava Barthakur, Srećko Joksimović and George Siemens write about how universities have turned to “short-term fixes such as ‘stress-testing’ written assessments and replacing them with oral examinations, hand written tests or reflective formats.” Such in-person assessments typically fall under the umbrella of ‘secure’ assessments. 

‘Secure assessments’ are those that use high levels of security to ensure that they are performed by the student unassisted by other supports, such as other people, outside source materials or Gen-AI. In some cases, secure assessments are required to meet program accreditation or institutional requirements (such as a requirement that all courses have a final exam). 

While in-person, proctored exams are not the only form of secure assessment, they are a popular choice. In today’s column, I invite you to reconsider the role of exams in your teaching. 

The challenges of secure assessments 

Secure assessments, and particularly proctored in-person exams, have at least three limitations: logistics, security and validity. 

Logistics:The logistical challenges of secure assessments are growing. As the University of Saskatchewan’s Gwenna Moss Centre for Teaching and Learning (GMCTL) writes, “assessments that need security often require more planning (e.g., technical configuration, physical materials, and space) and time restrictions. Conditions for assessment like these will likely also mean more [student access and equity] accommodations.” As many universities are experiencing significant increases in students approved for accommodations, secure assessments can lead to greater instructor and staff workload, as well as pressures on available secure accommodation assessment spaces. Online learning creates its own logistical challenges; among these, as the GMCTL identifies, are monitoring student actions during the exam to ensure the desired security; using browser lockdown software; and dealing with exam interruptions due to student loss of internet access.

Security:Exam security includes measures such as requiring students to display photo identification to verify identity, restricting students’ ability to leave the exam room, prohibiting personal items at exam desks, conducting sweeps of nearby areas to search for hidden cheat sheets, and so forth. While proctored in-person exams may be perceived as safe from AI-based cheating, wearable devices enable students to bring AI into the exam room. Discussing the technology of AI glasses, University of Calgary professor Susan Elaine Eaton and her coauthors write, “Models that feature a heads-up display project information onto the inside of the lens, which is not visible to an external proctor. This capability essentially allows a tudent to access the entirety of the internet and a powerful AI assistant during a high-stakes, closed-book exam, rendering traditional proctoring methods obsolete… The use of AI glasses lies beyond the scope of conventional proctoring methods, which are not designed to identify or regulate such discreet technology.” As University of Florida professor Sidney I. Dobrin writes, “trying to develop assignments for which GenAI platforms cannot provide viable responses may be impractical — if not impossible — given the velocity of AI evolution.” 

Validity: Hong Kong University of Science and Technology’s Center for Education Innovation director Sean McMinn suggests instructors ask two questions of our assessment practices: “What is the assessed task meant to prove?” and “Does this task still assess what I think it does?”

What are secure assessments meant to prove? Instructors are typically seeking to evaluate the student’s mastery of knowledge and/or skills independent of the assistance of another person, AI, or resource materials. But do secure assessments really assess mastery? Even without the security challenges noted above, the validity of these assessments can vary. Specific to exams, critics argue that the actual ‘proof’ of mastery is questionable: Longworth Education CEO Sarah Aiono argues, “Cognitive science tells us that knowledge retrieval is important for learning, but timed, high-stakes retrieval under exam conditions is a poor proxy for true understanding. A student may successfully memorise and reproduce information without being able to use or transfer it. Conversely, another student might deeply understand a topic but struggle to recall it under pressure or translate it into rapid written form.”Exam writing is a skill in itself, but probably not one that many educators prioritize as a learning outcome. 

Assessing unassisted learning: other options

While proctored exams are a common type of secure assessment, there are other options that instructors can consider.

Winona State University professor Steve M. Baule identifies oral assessments as a valuable approach: “Short, low-stakes oral defenses, whether one-on-one, in small groups, or recorded, create powerful validation opportunities. Students might: Summarize their key argument in two minutes; respond to clarifying questions; explain a specific data interpretation; justify a design decision. These conversations do not need to be high-pressure or time-intensive. Even a brief exchange can confirm whether the student understands the material.”  

Practical assessments provide another alternative. In this approach, students apply their learning to real-world tasks in a practical setting. This can include hands-on laboratory demonstrations, simulated scenarios, role playing, oral pitch or briefing presentations, live performances, or teaching materials to their classmates. 

Like proctored exams, these assessments support motivating students to master knowledge and/or skills and provide students with feedback about areas requiring further development. And like proctored exams, these assessments have logistical and security challenges that must be considered. 

Reconsidering ‘unassisted’ 

As we adapt to our brave new AI world, we need to start asking when it is truly imperative to have students demonstrate unassisted mastery of knowledge and/or skills. If, in their post-university lives, they will work in contexts of strong interpersonal and AI connection, when is the ability to work unassisted necessary? The point is not to eliminate all unassisted work or all exams, but rather to be thoughtful and deliberate about pedagogy.

Let’s use our current moment to reconsider our default assessment choices.