Over the coming years, new tools based on large language models (LLMs) and other artificial intelligence-based software are set to play an increasing role in many modern administrations, including in the anti-fraud domain. One might even argue that the prevention, detection, and investigation of fraud and associated illegal activities, which today involve processing and analysing an ever-growing volume of data of different types, are uniquely suited to the strengths of such tools. The authors of this article share some reflections on two particular challenges that authorities, which seek to harvest the potential of artificial intelligence for anti-fraud purposes, have to come to terms with: first, how to leverage the strength of artificial intelligence tools by identifying suitable use cases for the specific anti-fraud domain? Second, how to navigate the emerging regulatory framework considering in particular that the European Union’s Artificial Intelligence Act has entered into force on 1 August 2024?