The rapid expansion of generative artificial intelligence (AI) in higher education has intensified debates about its impact on learning, creativity and academic integrity. While early institutional responses have centred on plagiarism and assessment risks, emerging scholarship highlights a more urgent priority: the development of authentic, critically engagedAI literacythat enables students not only to operate AI tools but to interrogate their capabilities, limitations, and ethical implications. Enhancing AI literacy requires structured, experiential contexts in which students can experiment with AI, evaluate its outputs, and reflect on its appropriateness within disciplinary and professional contexts.

This paper introduces Sandboxed AI Learning, a pedagogical model designed explicitly to build such literacy by integrating AI purposefully and transparently within defined stages of the learning process. In this model, students may use AI during early exploration including ideation, research synthesis and visual experimentation, while AI‑generated content is strictly excluded from final assessed outputs. This sequencing positions AI as a catalyst for curiosity and conceptual expansion, without displacing the human criticality, ethical reasoning, and creative authorship that higher education seeks to develop.

The model is examined through Reimagining Road Safety, an authentic assessment in which undergraduate marketing students responded to a live brief from A Road Policing Unit Working across three interconnected modules, students developed personas, campaign concepts, and final communication materials while navigating real‑world constraints of accuracy and representation. Scaffolded stages, formative checkpoints and the deliberate withdrawal of AI support enabled students to experience AI’s potential, confront its limitations and then successfully move from exploratory AI‑assisted ideas into human‑authored outputs.

Findings demonstrate that Sandboxed AI Learning substantially enhances AI literacy by fostering evaluative judgement, ethical awareness and decision‑making. The model offers a replicable, responsible approach for institutions seeking to integrate AI meaningfully while preserving academic integrity and strengthening students’ readiness for an AI‑mediated world.