Large language models in STEM education: Opportunities, pedagogical shifts, ethical concerns, and future research directions

Large language models (LLMs) are rapidly transforming STEM and engineering education, yet their pedagogical value and responsible adoption depend on task design, systematic verification, and human oversight. This PRISMA-guided systematic review synthesizes evidence from 98 peer-reviewed studies published between 2021 and 2026, identified through a systematic search of major academic databases, to examine pedagogical applications, instructional transformations, ethical considerations, and research directions for LLM use in these disciplines. The findings indicate that LLMs can support a broad range of educational activities, including scalable tutoring, formative feedback, bilingual conceptual learning, programming, technical writing, assessment, and project- and capstone-based learning. However, the literature also highlights persistent challenges—academic integrity, output reliability, data privacy, algorithmic bias, intellectual property, and institutional governance—that underscore the need for carefully designed and appropriately governed integration. In the Saudi Arabian context, LLM adoption aligns with Vision 2030 and the National Strategy for Data and AI (NSDAI), yet realizing these benefits will require stronger institutional policies, faculty development, AI literacy, and assessment reform. The review proposes a multi-level conceptual framework that links national priorities, institutional governance, pedagogical integration, and student learning outcomes. Key research priorities include multimodal learning support, standardized prompt scaffolding, rubric-guided feedback, trustworthy tool-augmented systems, discipline-specific benchmarking, and longitudinal evaluation of educational outcomes. Overall, the review synthesizes current evidence to inform educators, policymakers, and researchers about the opportunities, challenges, and governance requirements associated with LLM integration in engineering education, while identifying priorities for future empirical validation.