资本市场人工智能使用监管工具包：独立工具包

SupervisoryToolkit for AIUsein CapitalMarkets:StandaloneToolkit The Board of theInternational Organization of SecuritiesCommissions Thisdocument is anextract of the full IOSCO reportSupervisory Toolkit for AIUse in Capital Markets. Itsupports IOSCO member authorities in theirrespective oversight of the use ofArtificial Intelligence (AI) based systems byentities subject to their regulation and supervision through a toolkit thatprovides supervisors with practical, non-binding, non-prescriptive supervisorytools, applicable across regulatory models. This report is the result of a multi-phased approach by IOSCO through its Fintech Task Force (FTF) to assistIOSCO members as they consider regulatory and supervisory responses to AItechnologies used in capital markets. This approach is based on a sharedunderstanding of the risks such technologies may pose to investorprotection, market integrity, and financial stability. Through extracting the toolkit from the full report into this standalonedocument, IOSCO aims toprovide supervisors with a practical reference thatcan be readily utilized during supervisory activities, including on-siteexaminations and inspections. By consolidating the tools in a single,accessible format, this document is designed to serve as a hands-onresource that supervisors can refer to when assessing supervised firms' useof AI systems. The full report providesbroadercontext on AI-relatedtechnological developments, risk analysis, and the principles of risk-basedsupervision that underpin this toolkit. These tools arenon-bindingand non-prescriptive and areinsteadintendedto be practicalandapplicable across different regulatory models. They arenot meant to be exhaustive but provide supervisors with a structured startingpoint and a flexible framework to identify risks and tailor their oversightapproaches to the specific characteristics andcontexts of AI deploymentwithin their markets. 1These are examples of indicators that may help monitor specific risks of AI usecases in each sector from the 2025 AI Report but are not exhaustive, nor are theyintended to be prescriptive. Data sources Tosupport oversight of financial institutions andcollection and analysis of theindicators highlighted above, supervisors may consider a variety of datasources, tools,and techniques. As highlighted in the FSB 2025 report2andcorroborated by the Survey responses, data is most often collectedthrough: •Supervisory reporting and data collection during examination andinspection:Data collected through supervisory reporting, examination,and inspection can include AI-related data where feasible. Such data,especially when collected on a regular basis, can enable quantitativeand qualitative analysis of AI use by supervised firms. •Regular and structured outreach and engagement with firms throughsupervisory engagement:Direct engagement with firms can enhanceunderstanding of how AI systems are being used in financial products and services and where risks may emerge, while at the same timereinforcing regulatory requirements and expectations. •Surveysof market participants:While surveys are used infrequentlybysupervisory authorities, according to the Survey results, sector-widesurveys can promote effective monitoring, covering AI inventories,third-partydependenciesandconcentrations,andAI-relatedinvestments, with survey designs that remain consistent yet adaptable,timely, and forward-looking to capture emergingdevelopments. •Research on publicly available information:Public reports (e.g.,annualreports,press releases,company disclosures, industry,and sectorreports) can provide information ahead of more focused engagement. •Real-time monitoring tools used by supervisors:Onlya few authoritiesreported employing real-time capabilities, including systems to analyzetradingbehavior through AI-powered detection tools.Real-timemonitoringimplies a cost-benefit assessment,but cross-borderinformationsharing can help reduce the costs and frictions indevelopingsuchcapabilities. •AItools:While most IOSCO members responding to the Surveyindicated that they are not currently using AI systemsto support theirmonitoring activities, a few do use AI systems for analyzing data fromroutine monitoring or public filings, to detectoutliers,anomalies,or toscanavailable datafor relevant information. Analyses reportedly arecarried out by using both commercial and internally developedtools. •Enhancedinformation sharing among national or cross-borderauthorities:Such information sharingcan help to minimize duplicationand improve coverage and comparability. •Sandboxes,innovation hubs,and other initiatives:Some IOSCOmembers are using other ways to gather information on AI use in firms,includingthrough engagement and outreach to industry,marketparticipants, and others.3