Supervisory screening with large language models: finding divergences

Key takeaways 

• We develop a procedure based on large language models to help supervisors review prospectuses against capital rules. This procedure ranks potential divergences for their consideration. We apply it to Additional Tier 1 (AT1) prospectuses from European global systemically important banks.
• We validate the procedure on anonymised documents from Credit Suisse and Yes Bank, recovering rule-prospectus differences that later mattered in disputes over AT1 loss absorption.
• This procedure could support other supervisory comparisons, including reviews of national rules against the international standards they implement.

The views expressed in this publication are those of the authors and do not necessarily reflect the views of the BIS or its member central banks.