BIS concludes Project Aurora, a proof of concept based on the use of data, technology and collaboration to combat money laundering across institutions and borders

ProjectAurora Thepowerofdata,technologyandcollaborationtocombatmoneylaunderingacrossinstitutionsandborders May2023 BISIHUnUnRrereesststrtPrriciicublctteteeidddc Preface:Howtoreadthisreport Foranoverviewoftheproject,usefulbackgroundandkeyfindingsread: Chapter2:Executivesummary Chapter3:Introduction Chapter4:ProjectAurora–proofofconcept Chapter6:Conclusion Formoresupportingdetailsread: Chapter5:Furtherconsiderations AnnexA:Trendsandopportunities Forotherdetailsread: AnnexesB-E ProjectAurorawasdeliveredinpartnershipwith: Publicationdate:May2023. ©BankforInternationalSettlements2023.Allrightsreserved.Briefexcerptsmaybereproducedortranslatedprovidedthesourceisstated. BISIHUnUnRrereesststrtPrriciicublctteteeidddc Contents 1.Acronyms,abbreviationsanddefinitions7 2.Executivesummary10 1.1Background10 1.2Data,technologyandinnovation11 1.3ProjectAurora12 1.4Findingsandkeytakeaways12 1.4.1Aholisticviewofpaymentsdataunveilsmoneylaunderingnetworks13 1.4.2Behaviouralmonitoringandprivacyenhancingtechnologiescould beagamechangerforAMLefforts 13 1.4.3LeveragingProjectAurora 14 3 Introduction 16 3.1Whatismoneylaundering? 16 3.2Themoneylaunderingprocess 17 3.3AMLmonitoringandanalysistoday 18 3.3.1Rule-basedmonitoringsystems 18 3.3.2Behaviouralmonitoringsystems 19 3.3.3Defensivereportingandde-riskingconsequences 19 3.4ChallengesfacingAMLefforts 20 3.5Technology 22 3.5.1Privacy-enhancingtechnologies(PETs) 22 3.5.2Graphdatastructures 23 3.5.3Machinelearning 23 3.5.4Networkanalysis 24 3.6Summaryoftrendsandopportunities 24 4. ProjectAurora–proofofconcept 26 4.1Objectivesandscope 26 4.1.1Objectives 26 4.1.2Scope 27 4.2PartA:Syntheticdatageneration 29 4.2.1Purposeofgeneratingsyntheticdata 29 4.2.2Generatingthesyntheticdata 30 4.2.3Leveragingathree-stepapproachtogeneratesyntheticdata 31 BISIHUnUnRrereesststrtPrriciicublctteteeidddc 4.2.4Constructingmoneylaunderingactivitiesinthesyntheticdataset34 4.3PartB:Applicationofmachinelearningtothesyntheticdataset39 4.3.1Machinelearningmodels39 4.3.2Testingthemodels41 4.3.3Results42 4.3.4Summary46 4.4PartC:Testingprivacy-enhancingtechnologiesforAML48 4.4.1Privacy-enhancingtechnologiesexplored48 4.4.2Testingacombinationofprivacy-enhancingtechnologiesinfourdifferentcollaborativeanalyticsandlearningarrangements49 4.4.3Results:applyingmachinelearningmodelsincombinationwithprivacy-enhancingtechnologies55 4.4.4PrivacyevaluationofPETswhenencryptingtransactiondata.61 4.4.5Summary62 5.Furtherconsiderations65 5.1Data65 5.1.1Additionaldataandmoneylaunderingtypologies65 5.1.2Real-worlddataarecrucialforunderstandingthefeasibilityandimpact66 5.1.3Limitationsofpaymentsdataandtheneedforotherdata66 5.1.4Dataprotection67 5.2Technology67 5.2.1TechnicalchallengeswithCALarrangements67 5.2.2Machinereadabletypologiesthatfacilitateinformationsharing68 5.2.3Explainability71 5.3Lookingahead71 5.3.1Instantpaymentsystems,CBDCsystemsandfinancialcrime71 5.3.2Legalandregulatoryconsiderations72 6.Conclusion74 7.AnnexA:Trendsandopportunities77 7.1Standardisation,transparencyandharmonisationinpayments77 7.1.1G20roadmapforenhancingcross-borderpayments77 7.1.2ISO20022harmonisation77 7.1.3Datastandardsforlegalentityidentificationandbeneficialownership78 7.1.4TheWolfsbergGroupPaymentTransparencyStandards79 BISIHUnUnRrereesststrtPrriciicublctteteeidddc 7.2Transactionmonitoringutilities79 7.2.1TMUexample:TransactionMonitoringNetherlands81 7.3InstantpaymentsystemsandpotentialCBDCsystems81 7.4Publicblockchainsusedforpayments82 8.AnnexB:MachinelearninginthisPoC83 8.1Machinelearningtraining,validationandevaluation83 8.2Machinelearningmodelfeatureengineering84 9.AnnexC:Privacy-enhancingtechnologies85 9.1OverviewofPETs85 9.1.1Homomorphicencryption85 9.1.2Localdifferentialprivacy85 9.1.3Federatedlearning86 9.1.4Otherprivacy-enhancingtechnologies86 9.2ApplicationofPETs87 10.AnnexD:Questionstosupportreal-worldpilots91 10.1Objectives,performancemonitoringandscope91 10.2Dataandanalysis91 10.3Post-pilotquestions93 11.AnnexE:Additionalacronymsanddefinitions94 12.References96 13.Acknowledgements100 BISIHUnUnRrereesststrtPrriciicublctteteeidddc 1.Acronyms,abbreviationsanddefinitions Theprojectspecificacronyms,abbreviationsanddefinitionsusedinProjectAuroraarelistedinbold.Otherusefulacronyms,abbreviationsanddefinitionsarelistedinAnnexE. AML Anti-moneylaunderingincludesallkindsofactions–includingsetsofrules,legislation,principles,regulations,processesandtoolsspecifictothefinancialsector–withtheobjectiveoftacklingthelaunderingofillicitlyobtainedfundsbycriminals. ANN Artificialneuralnetworksareasubsetofmachinelearning,inspiredbyasimplificationofneuronsinabrain.Theycanbeappliedtomodelcomplicatednetworkrelationshipsandpatterns. BIS BankforInternationalSettlements. CAL Collaborativeanalysisandlearning.Thiscollectivetermreferstoapproachesinwhichdifferentpartieseithercollaboratebysharingandanalysingdatainacentralisedmanner,orinwhichdifferentpartiescollaborateusingfederatedlearninginadecentralisedmannertotrainamachinelearningmodelonlocal