Risk Regulation of Generative Artificial Intelligence in the Australian Government: the Case of Microsoft Copilot

In many countries, risk regulation is central to AI regulation. We examine generative AI (genAI) risk regulation in Australia through a case study of the trial deployment of Microsoft Copilot in government agencies. Risk mitigation depended on end-users’ responsibility for human review and fact-checking, readiness testing, and contractual assurance from vendors, but largely ignored the impact on team dynamics and long-term implication on human abilities. Three areas of improvement were identified: strengthening fact-checking and review by users lacking in time, knowledge and experience; addressing the impact on team dynamics and human abilities; and measures for impending uses of genAI systems as internal and public-facing government chatbots which are anchored on government-mandated collaboration among developers, deployers and users.