From AI literacy to cognitive AI safety: rethinking safe AI use in education

Purpose This study aims to examine whether increasing AI literacy necessarily leads to safer and more responsible engagement with generative AI in higher education, and to introduce the concept of cognitive AI safety. Design/methodology/approach An explanatory sequential mixed-methods design was used. In Phase 1, survey data were collected from 311 university students to develop and test the cognitive AI safety scale (CAISS) and to examine whether AI literacy predicts cognitive AI safety. In Phase 2, an independent sample of 241 participants was used for confirmatory factor analysis. A qualitative phase then involved cognitive-audit interviews with 12 highly AI-literate participants to explore how verification, trust and decision-making were enacted in academic AI use. Findings The findings provide initial support for a revised three-factor structure of cognitive AI safety comprising epistemic vigilance, cognitive agency and failure resilience. Quantitative results show that AI literacy is positively associated with cognitive AI safety, though the model explained the relationship only modestly. The exploratory factor analysis supported the theorised three-dimensional structure, while the independent confirmatory factor analysis supported a revised 11-item version of the CAISS. Qualitative findings revealed that highly AI-literate users did not always apply verification. Instead, they checked selectively, guided by task stakes, domain familiarity, perceived effort and personal accountability. Originality/value This study introduces cognitive AI safety as a more focused, education-specific construct for understanding safe engagement with generative AI and provides initial validation of the CAISS. By integrating insights from AI literacy, epistemic vigilance, trust calibration and cognitive offloading, the authors reframe AI-related risk in education as a cognitive and pedagogical issue, as well as a technical and ethical one.