Seven lessons for AI-robust assessment design in higher education

The widespread adoption of AI by students in higher education has created urgent challenges for assessment design. As 89% of students use AI in their studies and nearly 80% use it for assessments, institutions are struggling to balance preventing misconduct with preparing students for responsible AI use. Existing reviews address parts of this challenge, but none comprehensively synthesize practical lessons for assessment design. This narrative review synthesized findings from recent empirical articles, previous reviews, and other recent studies into lessons that offer practical guidance to higher education professionals. Seven lessons are identified for AI-robust assessment design: (1) preventing cheating must not come at unacceptable costs to validity; (2) discursive solutions (e.g., AI use policies) must be combined with structural redesign; (3) a two-lane approach distinguishing independent from AI-augmented performance is advisable; (4) assessment design should protect the intrinsic cognitive load while allowing offloading of the extraneous load; (5) evaluative judgment has become a more central skill that requires extensive domain knowledge; (6) a diverse portfolio of assessment methods (the Swiss-cheese model) reduces vulnerability to both cheating and validity bias; and (7) AI-robust assessment requires sustained, integrated teacher professional development. Assessment design in times of AI is a wicked problem without quick fixes. The seven lessons offer a practical framework for institutions seeking to maintain validity, support learning, and uphold academic integrity in an AI-rich environment.