Speaker
Description
The rapid adoption of generative artificial intelligence (AI) is fundamentally reshaping language assessment by weakening the relationship between observed test performance and learners’ underlying language ability. This challenge is particularly significant in Vietnamese public security academies and universities, where VSTEP-style writing and speaking assessments play a high-stakes role in graduation, certification, and professional advancement. Despite growing concern about AI-assisted language production, current discussions have largely focused on AI detection and academic integrity, with limited attention paid to how generative AI challenges the theoretical foundations of assessment validity. Addressing this gap, this conceptual paper develops a framework for redesigning high-stakes writing and speaking assessments that remain valid in an era of ubiquitous AI access. Drawing on Messick’s unified theory of validity and Weir’s socio-cognitive framework, the paper employs a theory-informed conceptual analysis of recent scholarship on generative AI and language assessment. Rather than treating AI primarily as a problem of detection, the analysis reconceptualises it as a challenge to the evidential relationship between language ability and observed performance. With reference to assessment practices at the People's Security Academy as a representative context of high-stakes VSTEP implementation, the analysis identifies three interrelated threats to validity—cognitive processing, testing context, and score interpretation—and proposes a corresponding three-part redesign framework comprising process-visible task design, situated live performance, and evidence-oriented rating practices that prioritise demonstrable reasoning alongside linguistic performance. The resulting framework provides a theoretically grounded foundation for redesigning AI-resilient VSTEP-style assessments across Vietnamese public security academies and universities, which share similar assessment structures and high stakes attached to English certification, while also establishing a conceptual basis for future empirical validation and assessment innovation within public security education.
Keywords: generative AI; assessment validity; VSTEP; public security education; writing and speaking assessment
