Designing EFL Writing Assessment in the Age of Generative AI

Generative AI has rapidly reshaped how writing is taught and assessed in EFL contexts, raising concerns about assessment validity, student learning, and the limits of detectionbased responses. In Japanese university settings, where pressure for grammatical accuracy is common, unregulated AI use risks reducing opportunities for language development. This paper argues that these challenges are better addressed through assessment design than through prohibition or surveillance. Drawing on recent institutional guidelines and principles for ethical AI integration, it proposes a policyaligned, process-focused writing framework that differentiates permitted AI use across stages of the writing process. The framework combines AI-free in-class writing with clearly defined, learning-oriented AI use during revision, supported by transparent submission procedures.