Rethinking assessment for the AI era in practice, not in slogans
Universities are struggling to navigate the AI era and here’s why.
Generative AI has changed the game almost overnight. It’s not just about higher grades. It’s about what those grades signal and whether they still reflect the skills employers value. Research by Hausman, Rigbi and Weisburd (VoxEU, June 2025) shows a paradox: AI can boost performance, especially for weaker students, but it also compresses grade differences and risks eroding foundational skills. That leaves institutions asking: How do we maintain integrity, motivate learning, and prepare graduates for a world where AI is everywhere?
The challenge is real. Blanket bans don’t work they widen social gaps and ignore workplace realities. At the same time, over-reliance on AI can hollow out deep learning. Employers are already shifting focus from grades to portfolios, structured interviews and demonstrable competencies. Universities that cling to old models risk falling behind.
Who succeeds?
Those who rethink assessment design - not with slogans, but with deliberate strategy. They combine secure, performance-based exams with coursework that rewards ethical, critical AI use. They teach AI literacy as a core skill, broaden evidence of competence, and keep equity and motivation front and centre.
What we are seeing with universities that succeed is a blend of assessment formats: invigilated tasks that test what students can genuinely do under pressure, paired with takehome work that rewards transparent, critical and ethical use of AI. In digital exam terms, that means using secure conditions when needed and deliberately designing openbook activities where appropriate and making the expected use of AI explicit.
From our recent white papers, one pattern is striking: plagiarism hasn’t disappeared with generative AI, it’s diversified. Many cases are still “oldschool” (copying and paraphrasing, often from peers) and the grey zones around paraphrasing remain confusing for students. That’s why clear guidance and assessment design matter just as much as detection.
Five shifts institutions can make now
The recommendations below aren’t just theory they respond to a real tension universities face today. Generative AI is reshaping higher education faster than most institutions can adapt. As the VoxEU column “Generative AI in universities: Grades up, signals down, skills in flux” (Hausman, Rigbi & Weisburd, June 2025) highlights, AI can raise grades, especially for weaker students but it also blurs what grades signal and risks hollowing out core skills. Employers are already moving beyond transcripts, looking for portfolios and demonstrable competencies.
In short: the old assessment playbook no longer works. To stay relevant, universities need strategies that balance integrity, equity and motivation while embracing AI as a tool for learning not a shortcut. That’s where these five shifts come in. They are practical steps grounded in research and our own experience with WISEflow:
Modernize assessment formats. Combine supervised, performancebased tasks with coursework that explicitly credits good AI practice (prompt transparency, source citation, reflection on limitations). In WISEflow, secure exams and structured coursework can sit sidebyside in the same lifecycle. This balances the article’s finding that AI can raise grades (especially for lower performing students) with the need to evidence actual capability.
Teach AI literacy, don’t ban it. Blanket bans neither work nor prepare students for work. Equip students to use AI critically and ethically and assess those competencies. Our research shows confusion about rules is a key driver of misconduct; clarity and coaching reduce “accidental” cases and help mitigate the paper’s concern that grades may signal less about underlying skills.
Broaden the evidence of competence. If grade signals compress, employers will look for richer evidence: structured interviews, work samples, portfolios that document both fundamentals and AI augmented practice. Digital assessment workflows can capture and surface that evidence coherently over time.
Watch equity impacts closely. The study suggests AI lifts grades most for lower performers -npotentially narrowing gaps, but it may also displace key foundational skills. Policy should reduce inequalities (e.g., clear guidance, scaffolded tasks, feedback literacy) while still valuing core capabilities.
Strengthen intrinsic motivation. Overreliance on AI can blunt learning. Design for purpose, autonomy and mastery; make the why of each assessment visible, and provide timely, actionable feedback. This helps ensure gains in “AIspecific” human capital don’t come at the expense of deeper disciplinary skills.
To support this, we’ve embedded originality checking where it matters, inside the marking workflow. WISEflow Originality uses semantic similarity detection to surface paraphrasing and peer-to-peer copying across cohorts and sources, with the report available as an overlay during marking (no context switching). This is about faster judgement for markers and fairness for students.
And the reality check from our own work: a metaanalysis across 37 studies shows two dominant motivations behind deliberate plagiarism. 1) Reactive (time pressure, difficulty) and 2) Proactive (efficiency gains, low perceived risk)
Designing fair, motivating, aIaware assessments and detecting sophisticated similarity are two sides of the same integrity coin.
If you’re exploring how to align assessment design, AI literacy and integrity operations, our white paper series is a great next read. This series of four white papers goes deeper into Plagiarism in the age of AI.