美國《金融服務(wù)領(lǐng)域人工智能的用途、機(jī)遇和風(fēng)險(xiǎn)報(bào)告》

U.S.DEPARTMENTOFTHETREASURY

Uses,Opportunities,andRisksofArtificialIntelligenceinFinancialServices

ARTIFICIALINTELLIGENCEINFINANCIALSERVICES

REPORTONTHEUSES,OPPORTUNITIES,ANDRISKSOF

ARTIFICIALINTELLIGENCEINTHEFINANCIALSERVICESSECTOR

December2024

ExecutiveSummary

Uses,Opportunities,andRisksofArtificialIntelligenceinFinancialServices2

StaffAcknowledgements*

TheOfficeofFinancialInstitutions,includingActingAssistantSecretaryLaurieSchaffer,

DeputyAssistantSecretaryJeanetteQuick,DirectorMosesKim,leadauthorsSeniorPolicyAdvisorLiangJensenandSeniorPolicyAdvisorCaseyLaxton,andcontributorsSenior

PolicyAdvisorJoshNimmo,SeniorPolicyAdvisorBenHobbs,andFellowJackGinsberg,ledthepreparationofthisreport,incollaborationwithTreasurystafffromtheOfficeof

FinancialInstitutionsPolicy,FederalInsuranceOffice,OfficeofCapitalMarkets,OfficeofConsumerPolicy,OfficeofCyberSecurityandCriticalInfrastructureProtection,OfficeofGeneralCounsel,OfficeofInternationalFinancialMarkets,andOfficeofTerrorismand

FinancialIntelligence.

*Thecoverdesignofthisreportwasassistedbyartificialintelligence.

Contents

I.ExecutiveSummary 4

II.Background 7

III.KeyRecommendationsBasedonSummaryofResponses 10

IV.SummaryofResponsestotheAIRFI 14

A.ResponsesontheCurrentUsesandPotentialOpportunitiesofAIinFinancialServices 14

1.TraditionalAIandGenerativeAIUses 14

2.ExternalandInternalUsesbyFinancialFirms 15

B.PotentialRisksofAIandSuggestionsonRiskMitigations 16

1.DataPrivacy,Security,andQualityStandards 17

2.Bias,Explainability,andHallucinations 19

3.ImpactonConsumersandConsumerProtections 21

4.Concentration-relatedRisks 24

5.Third-PartyRisks 25

6.IllicitFinanceRisks 27

V.PolicyConsiderations 28

1.RegulatoryFrameworks 28

2.Federal,StateandOtherLegislativeEfforts 30

3.InternationalStandards 31

VI.PotentialNextSteps 32

VII.Appendix:Abbreviations 35

ExecutiveSummary

Uses,Opportunities,andRisksofArtificialIntelligenceinFinancialServices4

I.EXECUTIVESUMMARY

OnJune12,2024,theUnitedStatesDepartmentoftheTreasury(Treasury)

publishedarequestforinformationontheUses,Opportunities,andRisksofArtificial

Intelligence(AI)intheFinancialServicesSector(AIRFI).

1

ThroughtheAIRFI,TreasurysoughtinputonkeyissuesrelatingtoAIdeploymentwithinthefinancialservicessector,includingtheopportunitiesandriskspresentedtofinancialfirms,includingbanksand

nonbanks,bytheirownuseofAI,andtheopportunitiesandrisksfacingconsumers,

investors,businesses,regulators,end-users,andanyotherentityimpactedbydeploymentofAI.

2

TheAIRFIclosedforpubliccommentonAugust12,2024.InresponsetotheAIRFI,Treasuryreceived103commentlettersfromavarietyofstakeholders,includingfinancialfirms,consumeradvocacygroups,technologyproviders,financialtechnologycompanies,

tradeassociations,andconsultingfirms.Consideredintheirentirety,thesecomment

lettersdemonstratedthatAIisusedincreasinglythroughoutthefinancialsectortosupportabroadrangeoffunctionsandfirms.Therespondents

3

commentedonexistingusecases,expansiveopportunities,andassociatedrisks

4

,underscoringthepotentialforAIto

broadenopportunitieswhileamplifyingcertainrisks.Inparticular,manyrespondents

notedthatemergingAItechnologiessuchasGenerativeAIaredrivingexpandedusecasesbutalsointroducingnewrisk,leadingfirmstobecautiousaboutdeployingthembroadlyincustomer-facingapplications.Additionally,respondentshighlighteddifferencesin

supervisionforbanksandnonbanksdevelopinganddeployingAI,aswellastheresourcegapanddependencyonthird-partyprovidersforsmallerfinancialfirms.ThisfeedbackcarriesimportantimplicationsforfutureworkbyTreasuryandfinancialregulators,as

1TREASURY,REQUESTFORINFORMATIONONUSES,OPPORTUNITIES,ANDRISKSOFARTIFICIALINTELLIGENCEINTHEFINANCIALSERVICESSECTOR,89Fed.Reg.

50048(Jun.12,2024),

/documents/2024/06/12/2024-12336/request-for-information-on-uses-

opportunities-and-risks-of-arti?cial-intelligence-in-the-?nancial

(TreasuryAIRFI,ortheAIRFI).

2Id.“Financialinstitutions”intheAIRFIincludesbanks,creditunions,insurancecompanies,nonbank?nancialcompanies,?nancial

technologycompanies(alsoknownas?ntechcompanies),assetmanagers,broker-dealers,investmentadvisors,othersecuritiesand

derivativesmarketsparticipantsorintermediaries,moneytransmitters,andanyothercompanythatfacilitatesorprovides?nancial

productsorservicesundertheregulatoryauthorityofthefederal?nancialregulatorsandstate?nancialorsecuritiesregulators.This

reportusestheterm“?nancial?rms”insteadtore?ectthebroadscopeofinstitutionsincludedin?nancialservices.“Impactedentities”intheAIRFIincludesconsumers,investors,?nancialinstitutions,businesses,regulators,end-users,andanyotherentityimpactedby

?nancialinstitutions’useofAI.

3Throughoutthisreport,“respondent”isusedtoreferenceindividualsandentitiesthatprovidedpublicfeedbackthroughpubliccommentlettersabouttheAIRFI,aswellasthosewhoprovidedfeedbackdirectlytoTreasurysta?.

4Inadditiontotheriskshighlightedhere,respondentsalsoraisedconcernsaboutrisksrelatedtocybersecurityand?nancialstability.As

Treasuryauthoredorcontributedtoanalysisandrecommendationsrelatedtotheseareasinseparatematerials,theyarenotthefocusof

thediscussioninthisreport.See,e.g.,TREASURY,MANAGINGARTIFICIALINTELLIGENCE-SPECIFICCYBERSECURITYRISKSINTHEFINANCIALSERVICESSECTOR(Mar.2024),

/system/?les/136/Managing-Arti?cial-Intelligence-Speci?c-Cybersecurity-Risks-In-The-Financial-

Services-Sector.pdf

(TreasuryAICybersecurityReport);FINANCIALSTABILITYOVERSIGHTCOUNCIL,ANNUALREPORT(Dec.6,2024),

/system/?les/261/FSOC2024AnnualReport.pdf

(FSOC2024AnnualReport);FINANCIALSTABILITYOVERSIGHTCOUNCIL,ANNUALREPORT(2023),

/system/?les/261/FSOC2023AnnualReport.pdf

(FSOC2023AnnualReport);andFINANCIALSTABILITYBOARD,THEFINANCIALSTABILITYIMPLICATIONSOFARTIFICIALINTELLIGENCE(Nov.14,2024),

/uploads/P14112024.pdf.

ExecutiveSummary

Uses,Opportunities,andRisksofArtificialIntelligenceinFinancialServices5

existinglaws,regulations,andguidance,whileapplicableandlargelysupportedby

respondents,mayrequireenhancementtoaddressthegrowingcomplexitiesposedbyacceleratedAIdevelopment.Tothatend,manyrespondentsexpressedsupportforanumberofgovernmentactions,including:

1.AligningdefinitionsofAImodelsandsystemsapplicabletothefinancialservicessectortofacilitateinteragencycollaborationandcoordinationwithstakeholders;

2.Consideringprovidingadditionalclarificationonstandardsfordataprivacy,security,andqualityforfinancialfirmsdevelopinganddeployingAI;

3.Consideringexpandingconsumerprotectionstomitigateconsumerharm;

4.ConsideringclarifyinghowtoensureuniformcompliancewithcurrentconsumerprotectionlawsthatapplytoexistingandemergingtechnologiesandprovidingadditionalguidancetoassistfirmsastheyassessAImodelsandsystemsfor

compliance;

5.Enhancingexistingregulatoryframeworksanddevelopconsistentfederal-levelstandardstomitigaterisksassociatedwithpotentialregulatoryarbitrageand

conflictingstatelawswhileclarifyingsupervisoryexpectationsforfinancialfirmsdevelopinganddeployingAI;and

6.Facilitatingdomesticandinternationalcollaborationamonggovernments,

regulators,andthefinancialservicessectorandpursuepublic-privatepartnershipstoshareinformationandbestpractices,promoteconsistencyforstandards,and

monitorconcentrationrisk.

ThisreportprovidesbackgroundontheuseofAIinfinancialservicesbasedon

respondents’commentsandbuildingonobservationsfrompreviousTreasuryreportsandstakeholderengagement,

5

highlightsTreasury’songoingeffortstoevaluaterecent

developmentsinAI,andsummarizeskeyrecommendationsfromrespondentfeedback.

Next,thereportdetailstherespondents’commentsoncurrentandpotentialAIusecases,alongwiththeassociatedrisks,opportunities,andproposedriskmitigationstrategies.

Finally,thereportidentifiespolicyconsiderationsbasedonTreasury’sanalysisoftheAIRFIresponsesandlaysoutpotentialnextstepstobeconsideredbyTreasury,governmentagencies,andthefinancialservicessector,including:

1.Treasuryrecommendscontinuinginternationalanddomesticcollaborationamonggovernments,regulators,andthefinancialservicessectortopromoteconsistentandrobuststandardsforusesofAIinthefinancialservicessector.

2.Treasuryrecommendsfurtheranalysisandstakeholderengagementtoexploresolutionsforanyidentifiedgapsintheexistingregulatoryframeworks,

6

andto

5TREASURY,ASSESSINGTHEIMPACTOFNEWENTRANTNON-BANKFIRMSONCOMPETITIONINCONSUMERFINANCEMARKETS(Nov.2022),

/system/?les/136/Assessing-the-Impact-of-New-Entrant-Nonbank-Firms.pdf

(TreasuryNon-bankReport);TreasuryAICybersecurityReport,supranote4.

6FSOC2024AnnualReport,supranote4.Thereportnotedthattheauthoritytosupervisethird-partyserviceprovidersvariesamong

?nancialregulatorsandrecommendedthatCongresspasslegislationtoensurethatrelevantagencieshaveadequateexaminationandenforcementpowerstooverseethird-partyserviceprovidersthatinteractwiththeirregulatedentities.

ExecutiveSummary

Uses,Opportunities,andRisksofArtificialIntelligenceinFinancialServices6

addressthepotentialriskofAIcausingconsumerharm,asidentifiedbytherespondents.

3.Treasuryrecommendsfinancialregulatorscontinuecoordinatingtoidentify

potentialenhancementstoexistingriskmanagementframeworksandworkingwithothergovernmentagenciestoclarifysupervisoryexpectationsontheapplicationofframeworksandstandards,whereappropriate.

4.Treasuryrecommendsthefinancialservicessectorandgovernmentagencies

furtherfacilitatefinancialservices-specificAIinformationsharing,alongsidetheAIcybersecurityforumrecommendedintheTreasuryAICybersecurityReport,to

developdatastandards,shareriskmanagementbestpractices,andenhanceunderstandingofemergingAItechnologiesinfinancialservices.

5.TreasuryrecommendsthatfinancialfirmsprioritizetheirreviewofAIusecasesforcompliancewithexistinglawsandregulationsbeforedeploymentandthatthey

periodicallyreevaluatecomplianceasneeded.

Background

Uses,Opportunities,andRisksofArtificialIntelligenceinFinancialServices7

II.BACKGROUND

AIhaslongbeendeployedinthefinancialservicessectorinavarietyofways,

includingcreditunderwriting,insuranceunderwriting,trading,investmentadvice,

customerservice,compliance,forecasting,andprocessautomation.

7

TheuseofAIin

general–suchastraditionalmachinelearningalgorithms–canbetracedbacktothe

1940s.

8

ManyAIusesinthefinancialservicessectorcanbecategorizedunderthe

traditionalmachinelearningmethod–or“traditionalAI”–inwhichstatisticalmodelsare

trainedonadatasetwithinputandoutputparameters.

9

Thelasttwoyears,however,

markedamajorshiftfromtraditionalAIwithanaccelerationinthedevelopmentof

emergingAItechnologies–suchasdeeplearningmodelsutilizingneuralnetworksand

“GenerativeAI.”GenerativeAIdiffersfromtraditionalAIinitsabilitytocreatenewcontentbasedonwhatislearnedfromthetrainingdata.Generally,GenerativeAIreliesonmore

sophisticatedmodelsthataretrainedonvastamountsofdata.

10

ThePresident’sExecutiveOrder14110onSafe,Secure,andTrustworthyDevelopmentandUseofArtificial

IntelligencefromOctober30,2023(AIEO)

11

definedGenerativeAIas“theclassofAI

modelsthatemulatethestructureandcharacteristicsofinputdatainordertogeneratederivedsyntheticcontent”andnotedthatGenerativeAIcanproduce“images,videos,

audio,text,andotherdigitalcontent.”

12

FinancialfirmsareintheearlystagesofunderstandinganddeployingemergingAItechnology,includingGenerativeAI,asnotedintheAIRFIfeedback.EmergingAI,

includingGenerativeAImodels,aregenerallytrainedonmoreextensivedatasetsthan

traditionalAImodelsandrequiredistinctmodeldevelopmentprocessesandmodel

trainingtechniquescomparedtotraditionalAImodels.Manyfinancialfirmsrelyonthird-partiestodevelopanddeploythesemoreadvancedAImodels,andtherapidriseofopen-sourcetoolsisalsochangingthewaymodelsaredevelopedanddeployed.Additionally,

developingandoverseeingGenerativeAImodelsrequiresignificantlymoreadvancedexpertise,highercomputationalpower,andmoresubstantialfinancialinvestmentthantraditionalAImodels.Furthermore,GenerativeAImodelsexhibitagreaterlevelof

7SeeTHEALANTURINGINSTITUTE,ARTIFICIALINTELLIGENCEINFINANCE(Apr.2019),

https://www.turing.ac.uk/sites/default/?les/2019-

04/arti?cial_intelligence_in_?nance_-_turing_report_0.pdf;

TreasuryAIRFI,supranote1.

8SeeALANTURNINGINSTITUTE,supranote7;TreasuryAIRFI,supranote1.

9SARAHHAMMER,NAVIGATINGTHENEURALNETWORK:ARTIFICIALINTELLIGENCEINFINANCEANDRECALIBRATIONOFTHEREGULATORYFRAMEWORK(Oct.2024),

/wp-content/uploads/2024/10/Hammer-Navigating-the-Neural-Network.pdf.

10IBM,WhatAreFoundationModels?(Oct.11,2024),

/think/topics/foundation-models?utm_source=.

GenerativeAIsystemsarebuilton“foundationmodels”–atermthatdescribesatypeoflarge-scaleAImodeltrainedonvastamountsofdata,providingfoundationalcapabilitiestobe?ne-tunedforapplicationsacrossvarioussectors.

11WHITEHOUSE,EXECUTIVEORDER14110,SAFE,SECURE,ANDTRUSTWORTHYDEVELOPMENTANDUSEOFARTIFICIALINTELLIGENCE(Oct.30,2023),

/documents/2023/11/01/2023-24283/safe-secure-and-trustworthy-development-and-use-of-arti?cial-

intelligence

(AIEO).TheAIEOcallsforsociety-widee?ortthatincludesgovernment,theprivatesector,academia,andcivilsocietyto

meetthechallengesandopportunitiesposedbyAI.

12Id.

Background

Uses,Opportunities,andRisksofArtificialIntelligenceinFinancialServices8

complexityandcanbesusceptibleto“AIhallucinations,”

13

producingconfidentlystatedbutincorrectoutput,presentingnewchallengesforAIgovernanceandmanagement.

14

AstheuseofAIhasevolved,Treasuryhascontinuedtomonitortechnological

developmentsandtheirapplicationsandpotentialimpactsinfinancialservicestohelp

informanypotentialpolicydeliberationsoractions.TheAIRFIisoneofmanyeffortsin

whichTreasuryhasbeenengagingwithstakeholderstoimproveTreasury’sunderstandingoftheapplicationsofAIwithinthefinancialservicessector.Treasury’sNonbankReport

publishedin2022exploredopportunitiesandrisksrelatedtotheuseofAIinassessingtheimpactofnonbankfirms(includingfintechs)oncompetitionintheconsumerfinance

market.

15

AsdirectedbytheAIEO,TreasurypublishedareportinMarch2024onAIandcybersecurityandfraudrisksthroughextensiveoutreachonAI-relatedcybersecurityrisksinthefinancialservicessector.

16

InJune2024,theFinancialStabilityOversightCouncil(FSOC)Secretariathosteda

conferenceonAIandfinancialstabilitytoexplorepotentialsystemicrisksposedbyAIin

financialservices.

17

FSOCidentifiedtheuseofAIinfinancialservicesasasystemic

vulnerabilityinits2023AnnualReport.Inits2024AnnualReport,FSOCrecommended

continuedmonitoringoftherapiddevelopmentoftheusageofAItechnologiesinfinancialservicestoensurepoliciesareupdatedtoaddressemergingriskstothefinancialsystem

whilefacilitatingefficiency.

18

InSeptember2024,Treasuryhostedaroundtablediscussionwithrepresentativesfromtheinsuranceindustry,consumergroups,stateinsurance

regulators,academics,andotherstakeholderstogathermorefeedbackaboutAIintheinsurancesector.

19

Additionally,inMay2024,TreasuryissueditsNationalStrategyforCombating

TerroristandotherIllicitFinancing,notingthatAIhassignificantpotentialtostrengthenanti-moneylaundering/counteringthefinancingofterrorism(AML/CFT)compliancein

thefinancialsector.

20

InOctober2024,TreasuryreleaseditsinauguralU.S.National

StrategyforFinancialInclusion,whichacknowledgesthepotentialrisksandopportunitiesforAItoexpandconsumers’accesstofinancialproductsandservices.

21

13IBM,WhatareAIHallucinations?(Jun.2024),

/topics/ai-hallucinations.

14FLORENCEG’SELL,REGULATINGUNDERUNCERTAINTY:GOVERNANCEOPTIONSFORGENERATIVEAI(Sep.2024),

/content/regulating-under-uncertainty-governance-options-generative-ai.

15TreasuryNon-bankReport,supranote5.

16TreasuryAICybersecurityReport,supranote4.

17FINANCIALSTABILITYOVERSIGHTCOUNCIL,2024CONFERENCEONARTIFICIALINTELLIGENCEANDFINANCIALSTABILITY(Jun.6-7,2024),

/policy-issues/?nancial-markets-?nancial-institutions-and-?scal-service/?nancial-stability-oversight-

council/2024-conference-on-arti?cial-intelligence-?nancial-stability.

18SeeFSOC2024AnnualReport,supranote4;FSOC2023AnnualReport,supranote4.

19TREASURY,Readout:U.S.DepartmentoftheTreasuryHostsRoundtableonArti?cialIntelligenceintheInsuranceSector(Sept.24,2024),

/news/press-releases/jy2607.

20TREASURY,2024NATIONALSTRATEGYFORCOMBATINGTERRORISTANDOTHERILLICITFINANCING(May2024),

/system/?les/136/2024-Illicit-Finance-Strategy.pdf.

21SeeTREASURY,NATIONALSTRATEGYFORFINANCIALINCLUSIONINTHEUNITEDSTATES:FOSTERINGFINANCIALACCESS,RESILIENCE,ANDWELL-BEINGFORALL(Oct.2024),

/system/?les/136/NSFI.pdf.

Uses,Opportunities,andRisksofArtificialIntelligenceinFinancialServices9

Notably,Treasury’sOfficeofPaymentIntegritywithintheBureauoftheFiscal

ServiceannounceditslatesteffortsinenhancingfrauddetectionprocessesinOctober,includingtheuseofmachinelearningAItoexpeditetheidentificationofTreasurycheckfraud,whichresultedin$1billioninrecoveryoffraudandimproperpayments.

22

In

November2024,theOfficeofFinancialResearchreleaseditsannualreportwhichnotedthatAImayposenewrisksandvulnerabilitiesgivensuddenchangesinadoption.

23

Treasuryalsocontinuestoengagewithitsforeigncounterparts,includingthroughtheFinancialStabilityBoard,

24

G7,andG20,ontheimplicationsofAIfortheglobal

economyandfinancialstabilityandtopromoteinteroperabilityandalignmentinregulatoryapproaches.

Lastly,TreasuryhascontinuedtocoordinatewiththeFinancialandBanking

InformationInfrastructureCommittee(FBIIC)andtheFinancialServicesSector

CoordinatingCouncil(FSSCC)toestablishanAIExecutiveSteeringGroupandaddressissuesidentifiedintheTreasuryAICybersecurityReport.

22SeeTREASURY,PressRelease:TreasuryAnnouncesEnhancedFraudDetectionProcesses,IncludingMachineLearningAI,PreventedandRecoveredOver$4BillioninFiscalYear2024(Oct.17,2024),

/news/press-releases/jy2650.

23SeeOFFICEOFFINANCIALRESEARCH,ANNUALREPORT(2024),

https://www.?/annual-reports/?les/OFR-AR-2024_web.pdf.

24NELLIELIANG,U.S.TREASURYUNDERSECRETARYFORDOMESTICFINANCE,REMARKSONARTIFICIALINTELLIGENCEINFINANCE(Jun.4,2024),

/2024/06/remarks-by-nellie-liang-on-arti?cial-intelligence-in-?nance/.

KeyRecommendationsBasedonSummaryofResponses

Uses,Opportunities,andRisksofArtificialIntelligenceinFinancialServices10

III.KEYRECOMMENDATIONSBASEDONSUMMARYOFRESPONSES

Publicfeedbackfromrespondentsincludingabroadarrayofstakeholders,asoutlinedinSectionIVbelow,recommendedseveralpotentialactionstobetakenbygovernmentagenciesandthefinancialsector,including:

?AligndefinitionsofAImodelsandsystemsapplicabletothefinancialservicessectortofacilitateinteragencycollaborationandcoordinationwithstakeholders

Respondentsprovideddivergentviewsbutbroadlyagreedontheneedfor

alignmentondefinitionsofkeytermsandonthescopeofAIdefinitions.ManyrespondentssupportedthedefinitionintheAIRFIbutsuggestedTreasuryconsideralternative

definitions—suchasthoseusedbytheOrganizationforEconomicCo-operationand

Development(OECD)

25

andtheEuropeanUnion(EU).

26

Somerespondentsexpressed

concernsthatthedefinitionusedintheAIRFIwastoobroadandwouldthereforehinderinnovationsbyincludingtraditionalstatisticalmodelsandtechnologiesthatshouldnotberegulatedbyanAIgovernanceframework.Othersrecommendedadoptingawidely

accepteddefinitionofAIdevelopedbyastandard-settingbodytoensureclarity.Overall,respondentsgenerallyagreedthatconsistentAIdefinitionswouldenhancecollaborationamonggovernmentagencies,reduceregulatoryuncertainty,andminimizefrictionfor

financialfirmsoperatingacrossjurisdictions.

?Considerprovidingadditionalclarificationonstandardsfordataprivacy,security,andqualityforfinancialfirmsdevelopinganddeployingAI

Respondentsemphasizedtheneedofhigh-qualitydata–clean,complete,

standardized,andcomprehensive–fortrainingAImodels,testingefficacy,andreducingbias.Toensurehigh-qualitydatafordevelopinganddeployingAI,respondents

recommendedvariousstrategiestofinetunedatasetsanddatacuration.Ensuringdatasecurity–protectingstakeholders(includingfirms,consumers,andendusers)fromdata

25OECDde?nesan“AISystem”asamachine-basedsystemthat,forexplicitorimplicitobjectives,infers,fromtheinputitreceives,howto

generateoutputssuchaspredictions,content,recommendations,ordecisionsthatcanin?uencephysicalorvirtualenvironments.

Di?erentAIsystemsvaryintheirlevelsofautonomyandadaptivenessafterdeployment.OECD,OECDAIPRINCIPLESOVERVIEW,

https://oecd.ai/en/ai-principles.

26TheEUAIActde?nesan“AIsystem”asamachine-basedsystemthatisdesignedtooperatewithvaryinglevelsofautonomyandthatmayexhibitadaptivenessafterdeployment,andthat,forexplicitorimplicitobjectives,infers,fromtheinputitreceives,howtogenerate

outputssuchaspredictions,content,recommendations,ordecisionsthatcanin?uencephysicalorvirtualenvironments.EUAIACT,

https://arti?cialintelligenceact.eu/article/3/.

KeyRecommendationsBasedonSummaryofResponses

Uses,Opportunities,andRisksofArtificialIntelligenceinFinancialServices11

breachesanddatamanipulation–anddataprivacyiscriticaltorespondents.Respondentsalsoraisedconcernsaboutintellectualproperty,dataauthorization,andunauthorizeddatause.Somerespondentscalledforarobust,industry-widedataprotectionframeworkto

comprehensivelyaddresscurrentrisks,notingthattheexistingdataprotectionframeworkisfragmented.OtherrespondentssuggestedadoptingexistingorvoluntaryAIgovernance

andriskmanagementframeworkstoaddresstheserisks.Somerespondentsalsosuggestedlegislationthatencouragesprioritizingdataqualityandprotectionwhileavoidingcost-cuttingmeasuresthatunderminecriticaldata-relatedsafeguards.

?Considerexpandingconsumerprotectionstomitigateconsumerharm

RespondentsemphasizedtheneedforpolicymakersandregulatorstoaddressthepotentialriskofAImodelsandsystemscausingconsumerharmthroughopaquedata

collection,privacyviolations,andexacerbationofbiasesresultingindiscrimination.

Respondents’viewsvariedonwhetherexistinglaws–suchastheGramm-Leach-BlileyAct(GLBA)–aresufficientorneedenhancements.Manyrespondentssupportedanopt-in

modelfordatacollectionwithdrawaloptions,andregulatorylimitstoreducedatabreachrisks.Tomitigatebias-relatedrisks,respondentssuggestedincorporatingfairlending

principles,usingalternativedatafor“creditinvisible”consumers,andstrengthening

compliancemonitoring,whileurgingregulatorstoenhancefairlendingoversight,mandateexplainableAImodels,andpromotetransparency.

?ConsiderclarifyinghowtoensureuniformcompliancewithcurrentconsumerprotectionlawsthatapplytoexistingandemergingtechnologiesandprovidingadditionalguidancetoassistfirmsastheyassessAImodelsandsystemsfor

compliance

Whilefinancialfirmsmaybesubjecttoexistingconsumerprotectionlawssuchas

theFairHousingAct,EqualCreditOpportunityAct(ECOA)andFairCreditReportingAct

(FCRA),respondentssuggestedregulatorstakeadditionalstepstoclarifywaystoensure

uniformcompliancewithconsumerprotectionlawsthatapplytoexistingandemerging

technologiesbetweenbanksandnonbanksandbolsteringstrongercomplianceprocesses.Somerespondentsnotedthatadditionalregulatoryguidancecanencouragefirmsto

developanddeployAIinwaysthatleadtomorecompetitioninthemarket.Respondents

stressedtheneedtoaddressAIsystems’potentialtoexacerbatediscrimination.

Suggestionsincludedregulatorsmonitoringfinancialproductpricing,requiringexplainableAImodels,andidentifying“l(fā)essdiscriminatoryalternatives”–methodsthatachievethe

sameobjectiveswhilereducingbiasandthepossibilityofunfairoutcomes–forcompliancewithfairlendinglaws.

KeyRecommendationsBasedonSummaryofResponses

Uses,Opportunities,andRisksofArtificialIntelligenceinFinancialSer