Government agencies implement artificial intelligence (AI) systems to support or automate public services. While these systems can be beneficial, they may also yield adverse impacts, such as erroneous or unjust decisions on government benefits. Such impacts have triggered citizen activism and legal challenges, resulting in significant financial and reputational damages to governments worldwide. This project develops a novel theoretical model that advances science by shedding light on the anatomy of harmful AI systems in two ways. First, it provides a systemic explanation of how harmful government AI systems eventuate and exert impacts on citizens and other stakeholders. Second, it fills a critical void in knowledge on how harm caused by AI systems can be recognised and challenged. The model is examined via eight qualitative case studies of government AI systems from Europe and Australia. The project will generate new insights for responsible implementation and management of AI.

2025

2029