Governing Generative AI in Educational Assessment: Stakeholder Perspectives on Validity, Fairness, and Integrity in Turkey’s High-Stakes Testing

Abstract

There is a paradigm shift in educational measurement, as generative AI (GAI) impacts various aspects of test development, administration, scoring, security, and reporting. Meanwhile, it introduces new risks of validity, fairness, integrity, privacy and public trust. These risks are highest when the stakes are high and consequences follow formative assessments. This qualitative research study focuses on identifying the opportunities and threats that key stakeholders in the field of education measurement in Turkey see as a result of using AI. It also seeks to understand what governance conditions they see as essential conditions for responsible AI adoption in assessment. The research is conducted by a multi-stage design, which consists of document analysis and semi structured expert interviews, then analysed by reflexive thematic analysis. Five related themes were identified: purpose-first governance of AI in assessment; validity as construct integrity and evidentiary standards/documentation; fairness via ongoing monitoring of drift and subgroup effects; integrity and security considerations in the generative AI age, calling for new inference rules and assessment tasks; and legitimacy, privacy compliance and institutional readiness for sustainable use. The study finds that the responsible use of AI in assessment in Turkey requires context-based risk tiering, accountability, auditable documentation, continuous monitoring of fairness, integrity-by-design strategies and privacy-preserving governance. The results feed into the Responsible AI standards and the monitoring indicators for defensible, equitable and trustworthy assessment with the support of AI in Turkey.