AI on autopilot? A regulator’s case for automated analysis
Picture a banking supervisor arriving on Monday morning to find every new filing already analysed, peer comparisons run and outliers flagged before their coffee has cooled.
In the fifth instalment of his guest series, Björn Fastabend, head of the XBRL collection and processing unit at Germany’s Federal Financial Supervisory Authority (BaFin), argues that the technology to do this exists today.
His earlier pieces made the case for supervisors “chatting” with their data and verifying findings with business intelligence (BI) tools. Now he moves artificial intelligence (AI) upstream into the processing pipeline, where it runs on every report without anyone pressing a button. XBRL data was designed for automated processing, after all. There is a side benefit too: supervisors who would prefer to avoid new tools get the insights anyway.
Björn is candid about the pitfalls. Hallucinated false positives could be caught by a second “fact-check AI”, and any finding that leads to regulatory action needs a fully traceable audit trail. His proposed system pairs machine learning to spot what is statistically unusual with large language models that draw on the taxonomy to explain why it matters. As Björn puts it, “AI explores, BI verifies, humans decide”.
Read the full piece here.
