Do you copy? Attributing copyright infringement to actors involved in text-to-image generative artificial intelligence

Generative artificial intelligence (‘GenAI’) models democratise the artistic landscape by enabling amateurs and artists alike to create quality images. At the same time, copyright holders are increasingly concerned about copyright infringement where their works have been incorporated into training data without authorisation. We undertake a step-by-step analysis informed by technical information from recent scholarship and overseas litigation to consider the creation of training data, its use in machine learning, then the development and deployment of text-to-image GenAI models. This ‘supply chain’ approach allows us to assess the liability of actors at each stage, revealing the Australian legal position to be significantly more complex than has been appreciated. We explore the implications of this analysis, arguing in favour of a principled, rather than pragmatic, application of legal principles. We conclude by recommending reform agendas be animated by a broad view of copyright’s aims as well as an appreciation of its limits.