The article reframes the integration of generative AI in high-stakes English proficiency testing as a construct validity problem rather than merely an operational issue of security or scoring. It defines "AI-mediated construct drift" as the misalignment between test constructs anchored to unaided performance and target domain requirements increasingly involving AI assistance.
- Most existing literature treats AI only as assessment infrastructure, neglecting its implications for construct validity and extrapolation warrants.
- The paper proposes "bounded AI mediation" as a design principle where all test takers access the same institutionally controlled AI assistant with predefined boundaries and logged interactions.
- This approach aims to distinguish comprehension support from answer generation while maintaining standardized conditions.
The authors argue that score interpretations should be narrowed and supplemented when used to support claims about AI-mediated academic communication.