Beyond AI-Proofing: A Conceptual Framework for Assessment Design in the Age of Generative AI

Abstract

Though generative artificial intelligence now shapes many academic policies, institutions often react through bans, surveillance, and rigid rules. These measures fail at core levels, as they demand obedience to standards that lack practical enforceability, producing mere appearances of control instead of real educational integrity. This study suggests a different path under the name “Beyond AI-Proofing” that offers a restructured approach to assessment redesign built on four teaching and learning pillars. The first is validity: assessment should be about being confident that students can do what their degree says they can, not about searching for discrepancies that indicate potential academic dishonesty. The second is evaluating judgment: students need to learn to judge the quality of their work, other people’s work, and AI’s work, and this is something no machine can do on their behalf. The third is authentic process-focussed assessment: tasks should reflect real-world work and make the learning process visible, while being honest that authenticity is not a magic shield against AI. The fourth is a two-lane program structure: some assessments run securely in person, while the rest are open and AI is used openly with the whole program rather than any single task carrying the weight of proof. What ties these four together is that they build trustworthiness into the design of assessment itself, rather than trying to bolt it on through rules. Such a system holds up whether or not students use AI, because its integrity comes from how it is built.