从基础开始：盘点监督中的人工智能应用

Starting with the basics: a stocktakeof gen AI applications in supervision Jermy Prenio FSI Briefs are written by staff members of the Financial Stability Institute (FSI) of the Bank for InternationalSettlements (BIS), sometimes in cooperation with other experts. They are short notes on regulatory andsupervisory subjects of topical interest and are technical in character. The views expressed in them arethose of their authors and not necessarily the views of the BIS or the Basel-based standard setting bodies. Authorised by the Chair of the FSI, Fernando Restoy. This publication is available on the BIS website (www.bis.org). To contact the BIS Media and PublicRelationsteam,pleaseemailmedia@bis.org.Youcansignupforemailalertsatwww.bis.org/emailalerts.htm. ©Bank for International Settlements 2025. All rights reserved. Brief excerpts may be reproduced ortranslated provided the source is stated. Starting with the basics: a stocktake of gen AI applications insupervision1 Highlights •Many financial authorities are already experimenting with, developing or using generative artificialintelligence(gen AI) applications for supervision purposes.•Financial authorities seek to leverage the new technology to find information more efficiently, buttheir gen AI activities are hampered by outdated information technology (IT) infrastructure, datasecurity concerns and a lack of technical skills.•Most of the reported gen AI applications in supervision can be grouped into three categories: (i)basic document processing; (ii) knowledge management; and (iii) document review. Most “in use”applications fall into the first category; development work is spread out across the three categories;and experiments are concentrated in the second and third categories.•The main challenges identified in integrating gen AI applications in supervision are user acceptanceand inaccuracies in information provided. These challenges will likely intensify as financialauthorities move to more complex gen AI use cases. 1.Introduction The introduction of a generative artificial intelligence (gen AI) application in 2022 ushered in a flurry ofexploration and experimentation to harness the potential benefits of the technology in the financial sector.This has been aided by the accessibility of the new technology, with cloud service providers expandingtheir offerings to include gen AI applications. While financial institutions’ gen AI use cases focus onenhancing operational efficiency, they are continuously exploring other potential use cases and arereportedly investing heavily in AI in general.2 This exploration and experimentation around the potential benefits of gen AI are not confined tofinancial institutions. The development of gen AI also provides fodder for experiments by financial sectorsupervisors to leverage innovative technologies to aid supervision work, or so-called suptech. In general,supervisory authorities have always been interested in exploring applications that can help them gothrough huge volumes of narrative reports and other relevant text documents. Thus, over the years, therehave been many examples of tools using natural language processing. Gen AI has the potential tosignificantly improve the capabilities of these tools. Gen AI can also potentially make suptech applicationsmore accessible to users, thus addressing one of the challenges of effectively integrating such applicationsin supervisory processes.3 This paper provides an overview of the state of gen AI applications in financial supervision. It isbased on a survey of authorities represented in the Informal Suptech Network, which was established by the Financial Stability Institute (FSI) in 2018 as a forum for information exchange and sharing of practiceson how financial authorities integrate suptech in financial supervision. The survey was conducted fromJanuaryto March 2025,and responses were received from 42 authorities,4 with almost balancedrepresentation from advanced economies (AEs) and emerging market economies (EMEs). Respondentauthorities have varied mandates and include banking supervision, insurance supervision, anti-moneylaundering/countering the financing of terrorism (AML/CFT) and securities markets oversight (Graph 1).Section 2 outlines the gen AI activities of financial sector supervisors—ie whether they are experimentingwith, developing or already using gen AI applications. Section 3 describes the gen AI use cases insupervision, their perceived benefits, how they are deployed and the challenges faced in integrating themin supervisory processes. Section 4 concludes. Graph 1 Source: FSI. 2.Authorities’ gen AI activities Of the 42 respondents, 32 reported that they are already either experimenting with, developing or usinggen AI applications for supervision. Of these authorities, only 12 are already using such applications (Graph2). Comparing gen AI activities between authorities from AEs and EMEs, the survey responses show nomarked difference in the number