Framework for Governing GenAI Use in STEM Assessment

A recent study published on arXiv (2608.07475) introduces a framework aimed at students for managing the application of Generative Artificial Intelligence (GenAI) in STEM evaluations. This framework, based on Evidence Centered Design (ECD), outlines conditions for limiting, assisting, or mandating GenAI use. It tackles the governance issues arising from unrestricted access that may lead to task outsourcing and jeopardize assessment integrity, while outright bans are challenging to implement and could push usage underground. By offering decision rules that connect target constructs, evidence needs, and task attributes to governance strategies, the framework enhances existing AI use classifications. Restrictions are necessary when GenAI endangers relevant evidence for unaided proficiency, especially in fundamental knowledge and routine skills. Scaffolding is suitable when controlled GenAI assistance can improve learning without affecting validity. The study intends to equip students for future workplaces where GenAI-integrated processes are becoming more prevalent.