Grading in an age of assessment reform: the elephant in the room

Synthesis: Grades survived the first wave of GenAI assessment reform because both strategies on offer — proofing assessment against AI and automating its marking — leave the grade intact as the final representation of achievement. Nieminen and colleagues show universities redesigned tasks, feedback and outcomes while grading policy stayed silent, so reform stopped at the transcript. Changing that requires asking what achievement data is for, not which technology produces it.

This is a conceptual study with a policy review, not an empirical one: it reports no participants, interventions, or outcome measures. The university counts below come from the authors' review of publicly available policy documents.

Key Findings

1. Assessment reform is real but partial — tasks, feedback and learning outcomes are redesigned while the grade that certifies the result is left alone.
2. AI-proof grading defends the grade instead of rethinking it — invigilated exams, oral defenses and observed performances protect existing values and practices.
3. AI-powered grading reinforces the same system from the other side — efficiency and bias-reduction promises deepen the belief that achievement can be measured objectively.
4. Grading policy is silent — universities regulate how students and staff use GenAI, not how achievement is represented.
5. Grades persist because they are infrastructure — comparable numbers that travel between institutions, employers and the state.
6. The authors propose questions, not a replacement format — purpose, content, affordances, authorship and human-AI hybridity.

AI-proof grading protects the grade and the system around it

AI-proofing works at the task level. It covers invigilated pen-and-paper examinations, oral examinations answered in real time, and observed performance tests such as laboratory work. Cournoyea and Seeley's bot-proofing framework ranges from disallowing AI to punitive deterrence.

Keeping AI out of high-stakes tasks protects academic integrity and the public mandate that depends on trustworthy judgments. But the purpose is to safeguard the values and practices that grades already carry. AI-proofing is an act of stabilization: it entrenches the existing system and leaves its problems untouched. It can destabilize grades when applied to alternative forms of representation, such as supervised portfolio evidence, but those approaches are not at scale. The grade survives either way.

AI-powered grading improves the marker and leaves the grade untouched

AI-powered grading weaves automation into parts of the marking process while a teacher remains in the loop, promising efficiency at scale and less human bias. Both may materialize; neither challenges the grade.

Automated marking mimics what a human marker does, which routinizes grading and frames it as an administrative burden rather than a pedagogical judgment. The bias argument goes further: it grounds grading in a techno-rational paradigm that treats achievement as an objective quantity. An AI score can be more consistent than a tired marker and still sit inside the measurement logic that produced the grade. Accuracy is not the question: a better measurement of the same construct leaves the construct intact.

The policy review finds guidance for AI and silence about grades

The authors reviewed assessment and grading policies, plus GenAI policies, at universities in three systems in March 2026: 36 in Australia, 11 Finnish-speaking universities in Finland, and 8 in Hong Kong. They also consulted a review of 116 US research-intensive universities.

In Australia, all 36 universities provided GenAI assessment guidance and 13 had published official AI policies, yet none had renewed its grading policy. Most addressed assessment and feedback; some barred markers from using AI. In Finland, 9 of 11 universities had AI guidance, and none mentioned grades. All 8 Hong Kong universities had GenAI policies, and none addressed changes to letter grading. The US review found privacy advice and some encouragement to use GenAI for grading, but no institutional redefinition of achievement.

Policy reform takes time, and public documents cannot capture faculties, programs or classrooms. Still, the redesign rhetoric that pervades assessment policy is absent from grading policy. Grades remain the elephant in the room.

Reform would have to ask what grades are for

Grades are not only symbols. They are a network of relations: students come to see themselves through them (learner identity), the teaching relationship is constituted by grading, and the numbers travel into scholarships, exchange programs, doctoral admissions and labor markets. That portability keeps grades in place, and explains why assessment reform can leave the transcript untouched.

Removing grades is not a solution by itself. Pass/fail grading can create new forms of competition, and a changed scale may leave the underlying meaning intact. The authors instead propose five questions: For what purposes and for whom is achievement data produced? What achievement should be represented institutionally, and what should not? What do different forms of data afford? Who constructs the data, and can students have a genuine role in representing their achievement? How can representations account for human-AI hybridity, including whether and how tools were used? They add an equity demand: replacements should not flatten the diverse knowledge resources students bring.

What this means for practice

• Instructors: When you AI-proof a task, state what the resulting grade will and will not tell a reader. A defended exam protects the mark; it does not make it a fuller account of what the student can do.
• Program leaders: Check whether program-level reform changed anything a transcript records. If not, it is local to tasks while the program certifies achievement the old way.
• Assessment designers: Treat AI grading as a measurement decision before a workflow decision: document what the marker claims to measure, where human judgment stays accountable, and what students are told.
• Administrators: Write the grading question into AI policy. Guidance on student use and marking is not a position on whether letter grades remain the right representation.
• Policymakers: Ask institutions to justify the achievement data they publish — who it serves and what it cannot show — not only whether AI produced it.

Limitations

• This is a conceptual argument, not a test of it: the claim that both responses reinforce the status quo is interpretive.
• The policy review covers public documents in three systems at one time, so faculty, program and classroom variation remains invisible.
• All three systems are well resourced; the US evidence is second-hand, and no low- or middle-income system is examined.
• The literature on AI grading is moving quickly, and the review is illustrative rather than systematic; the paper does not evaluate whether alternatives such as programmatic assessment work at scale.

Citation

Nieminen, J. H., Bearman, M., Boud, D., Corbin, T., Dawson, P., Tai, J., & Walton, J. (2026). Grading in an age of assessment reform: the elephant in the room. Assessment & Evaluation in Higher Education. https://doi.org/10.1080/02602938.2026.2694451