Validity Architecture for AI Integrated Assessment (VAAI): A Conceptual Framework for Defensible Assessment Practice

Generative AI has weakened the link between submitted work and warranted claims about learner competence, creating an inference problem that detection and prohibition alone cannot resolve. This article introduces the Validity Architecture for AI Integrated Assessment (VAAI), a conceptual framework for redesigning assessment when AI assistance may shape student work. Grounded in argument-based validity and construct validity theory, VAAI treats AI-related assessment challenges as problems of inference, evidence, and interpretation rather than primarily as problems of academic integrity. For practitioners, it organizes assessment design around five questions: what claim is being made, what evidence can support it, what AI assistance is compatible with the claim, how scoring and fairness should be handled, and when the assessment must be reviewed. VAAI combines a maturity scale across eight dimensions with an evidence model centred on provenance, adversarial rebuttal testing, and consequence-calibrated implementation. Worked examples from nursing, counselling, and program-level capstone review illustrate how the same validity logic applies across disciplines and consequence levels. The article contributes a structured validity architecture for making interpretive claims, assistance boundaries, and evidentiary requirements more explicit, contestable, and defensible in higher education assessment practice.