Shadow AI learning in organisations: a sociomaterial framework for employee-driven artificial intelligence training and knowledge management

Purpose This paper aims to develop a theoretical framework explaining how informal, employee-driven artificial intelligence (AI) learning contributes to organisational knowledge management through a sociomaterial lens. It addresses a gap in knowledge management theory by treating grassroots AI experimentation as a legitimate form of organisational learning and knowledge creation. Design/methodology/approach This conceptual paper uses a theory synthesis approach (Jaakkola, 2020) to integrate sociomateriality with informal learning theory. Following established methods for rigorous conceptual research (Heinonen and Gruen, 2024; Meredith, 1993), this paper develops the shadow AI learning framework, including six testable propositions spanning micro, meso and macro-organisational levels. Findings The shadow AI learning framework shows that employee-driven AI experimentation generates valuable organisational knowledge through sociomaterial entanglement across all three levels. The framework traces how informal learning moves from individual discovery to collective capability, moderated by organisational culture, leadership attitudes and quality assurance mechanisms. Originality/value The primary contribution is the identification of three mechanisms through which shadow AI learning reconfigures conventional informal learning: expertise-independent transferability, in which knowledge artefacts carry embedded experimentation to recipients who did not produce them; learning cycle compression, in which iterative querying accelerates the externalisation of tacit insight; and semi-stable cognitive infrastructure, in which accumulated artefacts persist as organisational resources beyond their creators. The framework positions shadow AI learning as a new configuration of established processes, not a wholly distinct phenomenon.