Generative artificial intelligence (GenAI) is rapidly permeating science, technology, engineering, and mathematics (STEM) education across the board—from learners, to institutions, to education policy. While GenAI tools promise benefits such as scalable feedback, personalized guidance, and automation of instructional tasks, their widespread adoption also raises critical concerns. In this commentary, we argue that many current GenAI applications conflate learning with performance and feedback, neglecting the active, reflective, and embodied processes that underpin meaningful learning in STEM disciplines. We identify three core challenges: (I) authentic learning requires active integration of knowledge, but (II) existing GenAI tools are not designed to support such processes, and thus (III) uncritical use of GenAI may undermine the very goals of STEM education by fostering (meta-)cognitive debt, de-skilling, and misplaced trust in machine-generated authority. We further highlight the systemic risks of commercial lock-in, epistemic opacity, and the erosion of educational institutions’ role in cultivating critical engagement with subject matter. Rather than assuming that access to GenAI equates to improved learning, we call for rigorous, discipline-based research to establish under which conditions GenAI can meaningfully contribute to STEM education. The central question is not whether GenAI will be adopted, but when, how, and why it should be used to enhance, rather than diminish, opportunities for active and reflective learning.
