Marco Schreyer from the SFAO, in collaboration with researchers from the Deutsche Bundesbank and the University of St.Gallen, presents a novel approach for irregularity detection in tabular data. In doing so, he combines diffusion models with autoencoders to significantly improve the detection of irregularities in financial and audit data. Results demonstrate improvements of up to 65% compared to conventional methods, opening new possibilities for digital auditing and fraud detection.