Artificial intelligence is increasingly being used in statistical production, but its responsible adoption in Official Statistics requires more than technical experimentation. Drawing on ongoing work at Istat, this presentation discusses selected use cases that highlight the methodological and institutional challenges of using AI-based models in statistical processes, including validation, quality assessment, transparency, data governance, and confidentiality protection. The presentation will show how Istat is addressing these issues to support the transition of AI-based methods and products into production environments while preserving statistical quality, institutional accountability, and public trust.
