Unapproved Generative AI Use (Shadow AI) in the Digital Workplaceand Study Contexts : A Quantitative Study of Bypass Behaviour

Sammanfattning

Generative artificial intelligence is increasingly used in work and study contexts, but its use may occur through tools, accounts, or services that are not officially approved by an organisation or university. This thesis examines such use as Shadow AI, defined as reported use of unapproved generative artificial intelligence tools or accounts for work or study tasks. The study conceptualises Shadow AI as non-malicious, task-oriented bypass behaviour rather than intentional harm. While workarounds and Shadow IT are established topics in Information Systems research, less is known about how these explanations apply to unapproved generative AI use in everyday work and study contexts. Drawing on Information Systems research on workarounds, Shadow IT, perceived usefulness, quantitative workload, information security fatigue, and policy malabsorption, the thesis examines whether four factors are associated with Shadow AI use: perceived usefulness, quantitative workload, information security fatigue, and generative artificial intelligence policy malabsorption. A quantitative cross-sectional survey was conducted. The final dataset contained 196 consented responses, with 176 respondents included in the main valid analysis sample after screening and attention-check rules. Shadow AI use was operationalised through two direct outcomes: Shadow AI bypass and unapproved-use intensity. The findings show that Shadow AI use was present in the sample. The clearest empirical finding concerned GenAI policy malabsorption among respondents who were aware of institutional GenAI policy. In that policy-aware model, respondents who found policy harder to absorb were more likely to report Shadow AI bypass. The primary model was significant overall, but no individual predictor reached the conventional significance threshold, so the results should be read as cautious association-based evidence.