GenAIRisksforBusinesses:Exploringtheroleforinsurance

October2025

ASSOCIATION

G\GENEVA

GenAIRisksforBusinesses:

Exploringtheroleforinsurance

Ruo(Alex)Jia

DirectorDigitalTechnologies,GenevaAssociation

AssociateProfessorofInsurance,PekingUniversity

Contributingauthors:

MartinEling

DirectoroftheInstituteofInsuranceEconomicsand

ProfessorofInsuranceManagement,UniversityofSt.Gallen

TianyangWang

ProfessorofFinance,ColoradoStateUniversity

1

GenevaAssociation

TheGenevaAssociationwascreatedin1973andistheonlyglobalassociationof

insurancecompanies;ourmembersareinsuranceandreinsuranceChiefExecutive

Officers(CEOs).Basedonrigorousresearchconductedincollaborationwithour

members,academicinstitutionsandmultilateralorganisations,ourmissionisto

identifyandinvestigatekeytrendsthatarelikelytoshapeorimpacttheinsurance

industryinthefuture,highlightingwhatisatstakefortheindustry;develop

recommendationsfortheindustryandforpolicymakers;provideaplatformtoour

membersandotherstakeholderstodiscussthesetrendsandrecommendations;

andreachouttoglobalopinionleadersandinfluentialorganisationstohighlight

thepositivecontributionsofinsurancetobetterunderstandingrisksandtobuilding

resilientandprosperouseconomiesandsocieties,andthusamoresustainableworld.

Photocredits:

Coverpage–Unsplash

GenevaAssociationpublications:

PamelaCorn,DirectorCommunications

HannahDean,Editor&ContentManager

JooinShin,DigitalContent&DesignManager

Suggestedcitation:GenevaAssociation.2025.

GenAIRisksforBusinesses:Exploringtheroleforinsurance.

Author:Ruo(Alex)Jia;Contributingauthors:MartinEling;TianyangWang.October.

?GenevaAssociation,2025Allrightsreserved

2

Contents

Acknowledgements4

Foreword5

Executivesummary6

1.Introduction8

1.1GenerativeAI:Definitionandapplications9

1.2Gen-AI-inducedrisks11

1.3Researchquestionandcontributions16

2.DemandforGen-AI-relatedinsurance:Abusinesscustomersurvey17

2.1ApplicationofGenAIinbusiness19

2.2AwarenessandperceptionofGenAIrisks21

2.3DemandforGen-AI-relatedinsurance22

3.SupplyofGen-AI-relatedinsurance26

3.1InsurabilityofGen-AI-relatedrisks27

3.2Emerginginsurancesolutions32

4.Conclusion,outlook,andrecommendations36

Appendix1:ExistingAIriskclassifications40

Appendix2:InsurabilityofGenAIrisks(extended)41

References43

3

4

ACKNOWLEDGEMENTS

ThisreportwaspreparedundertheguidanceoftheGenevaAssociation’sDigitalTechnologiesworkinggroup,sponsoredbyBiancaTetteroo,ChairoftheExecutiveBoardofAchmea.

Wearegratefultothefollowingexpertsandexecutiveswhomadethemselvesavailableforinterviewsorprovidededitorialcontributions:

?TomoAsaka(TokioMarine)

?BartBoonandRenéWissing(Achmea)

?ColonnellaEmanuele(EdgeGroup)

?SophieFarhaneandEléonoreJacquemin(AXA)

?MatthewGabriel(Manulife)

?JesusGonzalez(Aon)

?AtsushiIzu(Dai-ichiLife)

?ChristophKrieg(Peak3)

?YunlongLIU(PICC)

?PaulLloydandRobertMilanPorsch(AIAGroup)

?HugoSantaMaria(Fidelidade)

?LamiaElMarzouki(AXAXL)

?DennisNoordhoek,DarrenL.Pain,andKai-UweSchanz(GenevaAssociation)

?FrankSchmid(GenRe)

?JoanSchmit(UniversityofWisconsin-Madison)

?BillSchwegler(Transamerica)

?SimonTorrance(EmbeddedFinance&InsuranceStrategies)

?JingXIAO(PingAn)

ThereportalsobenefittedfromdiscussionsattheGenevaAssociation’s2024ProgrammeonRegulationandSupervision(PROGRES)and2024DigitalTechnologiesConference.

Finally,weextendourdeepestthankstotheGenevaAssociation’sEditorialCommitteeandAssociatesfortheirsupportandfeedback,aswellastoPieralbertoTreccani(formerlyoftheGenevaAssociation)andXinyuFAN,QinyuLI,andYunfeiYANG(PekingUniversity)fortheirvaluableresearchassistance.

Foreword

Inmomentsofprofoundtechnologicaltransformation,weoftenfaceaparadox:thetoolsthatpromisetoelevateuscanalsochallengeusinunexpectedways.GenerativeAI

demandsbothstrategicanticipationandhuman-centredstewardship.

Thisreportisourcontributiontoafast-evolvingconversationabouthowto

understand,manage,andinsureagainstthenewclassesofriskthatGenAIcreatesoramplifies.Fromcybersecuritythreatstointellectualpropertychallenges,from

workforcedisruptiontoliabilityuncertainty,arisklandscapethatgoesbeyondtraditionalcategorisationisemerging.

Togroundourinsightsinreal-worldexperience,weconductedaglobalsurveyof600businessrepresentativesinvolvedincorporateinsurancedecision-making,acrossthelargestsixinsurancemarkets.Oneclearfindingstandsout:morethan90%of

respondentsseeaneedforinsurancecoverageforGenAIrisks,withtwothirdswillingtopaymoreinpremiumsforit.Thissignalsbothurgencyandopportunity.

TheinsuranceindustryhasacriticalroletoplayinofferingprotectionandinshapingresponsibleGenAIadoption.Byapproachingthesechallengescollaboratively–withtechnologyproviders,regulators,andbusinessesalike–andwithcaution,wecan

helpensurethatthebenefitsofGenAIarerealisedsafelyandsustainably.

JadAriss

ManagingDirector

5

6

Executivesummary

AdoptionofGenAIheightensoperational,ethical,andcybersecurityrisks,spurringdemandfor

insuranceamongbusinesses.

BusinessesarerapidlyintegratingGenerativeAI(GenAI)

intobothcustomer-facingproductsandservicesand

theirinternaloperations.Thisintroducesnewriskssuchasdefectiveoutputs,biasedrecommendations,intellec-tualpropertyinfringements,andcybersecurityconcerns.TheserisksbecomeparticularlyprominentwhenGenAImodelshallucinateorreplicateprotectedcontent.

GenAIintroducesbothbenefitsandriskstobusinesses.

ThisreportexploresGen-AI-relatedrisksandassessesbusinesses’awarenessanddemandforrelated

insurance.Buildingonestablishedframeworks,we

classifyGenAIrisksintosevendomains:operational,

cybersecurity&privacy,ethical,regulatory,reputational,workforce,andESG.TheseemphasisehowGenAI

amplifiesorcreatesexposuresbeyondtraditionalriskcategories.

ToevaluateriskawarenessandinsurancedemandfrombusinessesthatuseGenAI,theGenevaAssociation

commissionedasurveyof600corporateinsurance

decision-makers/influencersacrossthesixlargest

insurancemarkets(China,France,Germany,Japan,theUK,andtheUS).ThesurveyresultsrevealwidespreadGenAIadoption,thoughperceivedusefulnessvaries

byregion–itishighestinChinaandtheUS–reflectingdifferinglevelsofdigitalmaturityandorganisational

culture.

BusinessesfacesignificantGenAIimplementation

hurdles,particularlytalentshortages,poordataquality,andinternalresistance.Theprimarychallengesvary

acrossmarkets,influencedbydifferinglevelsofwilling-nesstoadoptGenAI.

Cybersecurityrisksemergeasthetopconcernofbusi-nesses,citedbyoverhalfofsurveyedfirms,followed

bythird-partyliabilitiestoclientsandsuppliersandthenoperationaldisruption.Reputationaldamagerankslowerdespiteitspotentialforlong-termimpact.

Morethan90%ofrespondentsexpressaneedfor

insurancecoveragetailoredtoAI/GenAIthreats;over

twothirdswouldpayatleast10%moreinpremiums

forexplicitinsurancepolicyextensionsthatcover

Gen/AIrelatedrisks.Demandisparticularlystrong

amongmediumandlargeenterprises,inthetechnologyandfinancesectors,andinregionswithhigherGenAIadoption.Additionally,highGenAIriskexposureand

highseverityofpastGenAIfailuresdriveinsurancedemand,suggestingpotentialadverseselection.

Demandforinsurancethatcovers

GenAIrisksishigh,particularlyamongmediumandlargefirmsandinthe

technologyandfinancesectors.

Onthesupplyside,applyingBerliner’sinsurability

frameworkrevealsinsurabilitychallenges,atleastin

theshortterm.GenAIrisksmayleadtolargepotentiallosses.AsitisdifficultforinsurerstoverifyGenAIrisksandhowbusinessesmanagethem,Gen-AI-related

insurancemayexperienceseriousinformationasym-

metry.Insurersmaythereforebereluctanttoofferhighcoveragelimits,asintheearlydaysofcyberinsurance.

InsurersarerespondingtoGenAIrisksbyadapting

cyberandliabilitypoliciestoincludeGen-AI-related

causesofloss;parametrictriggersanddue-diligence

protocolsarebeingtestedtostreamlineunderwriting

andclaimsprocesses;andselectedstandaloneAIinsur-ancesolutionsthatintegratevarioustypesofcoverageintoasinglepolicyindicatetheemergenceofanascent

7

market,thoughitremainstooearlytosaywhether

existinginsuranceproductsornewstandalonesolutionswillcometodominatetheGenAIriskmarket.

InsurersareadaptingcyberandliabilitypoliciestoincludeGen-AI-relatedrisks,whilestandalonecoverageisalso

emerging.

TokeeppacewithGenAIinnovation,insurersshould

proactivelydefineGenAI’sriskboundariesandbegin

pilotingmodularcoverageextensions,beforeloss

eventsforcereactiveresponses.Insurersmayconsiderpartneringwithtechnologyprovidersandregulatorstoco-developriskassessmentframeworksforGenAI,

embedcontinuousmonitoringinpolicyterms,and

exploresimulation-basedmodelling.Suchcollaborationwouldharmoniseethicalstandards,clarifycoverage

terms,andstrengthentheinsuranceindustry’srole

insafeguardingandsupportingthedevelopmentandadoptionofGenAI.

1Introduction

9

Introduction

GenAIpromisestransformativegainsinproductivityandcreativity,yetitsopacityandautonomyintroduceriskswithfew

historicalparallels.

1.1GenerativeAI:Definitionandapplications

GenerativeAI(GenAI),isasubsetofartificialintelli-

gencethatcancreateoriginalcontentsuchastext,

images,voices,videos,andtheircombinationsin

responsetouserrequests.1GenAIisarevolutionary

digitaltechnologythathasthepotentialtofundamentallyreshapeproductionprocessesineconomies,much

likeearlierbreakthroughssuchasthesteamengine,electricity,andtheinternet.

GenAIisarevolutionarydigital

technologythathasthepotentialtofundamentallyreshapeproductionacrosseconomies.

GenAIbuildsonmanyofthestatisticaladvances

underpinningtraditionalAI(seeBox1).InGenAImodels–especiallylargelanguagemodels(LLMs)–thecore

taskistopredictthenexttoken(e.g.awordorphrase)inasequence,giventhecontextofallprevioustokens.Thissequentialtoken-by-tokenpredictionenables

theGenAImodeltogeneratecoherent,contextually

relevanttextorothercontentthatappearsfluidand

human-like.TraditionalAItypicallyinvolvestraditionalmachinelearningtechniquessuchasclassificationorregression,wherethemodelistrainedtomappotentialpredictorsdirectlytoafixedtargetvariable(e.g.fore-castingtomorrow’stemperature).

Box1:GenAIvs.traditionalAI

GenAIinvolvespredictingthenexttokeninasequence.Thisprocessinvolvesoptimisinganobjectivefunction,whichguidesthemodelingeneratingcoherentand

contextuallyrelevanttext.Commonobjectivefunctionsincludemaximisingthelikelihoodofthenexttokengiventheprecedingsequenceorminimisingthedifference

betweengeneratedandtargetsequences.Optimisationtechniqueslikegradientdescentareusedtofine-tunethemodel’sparameterstoachievethisobjective.The

enginebehindGenAIisdeeplearning,anadvancedtypeofmachinelearningbasedonneuralnetworks,whichcanprocessunstructureddataandextract

featuresfromdataautomatically.3

GenAIdiffersfromtraditionalAIinitsabilitytocreate

entirelynewcontentratherthanmerelyanalysingdataandmakingpredictionsbasedonpre-existingpatterns.2

TraditionalAIlearnsfromextensivedatasetstoidentifypatterns.Itsprimarystrengthliesinprocessingstruc-tureddata.TraditionalAIisclosertoclassicalstatisticalmodels,wherebythereisamathematicalexpressionthatquantifiestheperformanceofamodelandguidestheoptimisationprocess.Machinelearningalgorithmsareusedtoadjustthemodel’sparameterstooptimisetheobjectivefunction.

Source:GenevaAssociation

1

IBM2024a

.

2

HermannandPuntoni2024

.

3

Ramakrishnan2025

;

GenRe2025

.

10

InNovember2022,OpenAIlaunchedChatGPT,a

conversationalAItoolthatrapidlygainedtractionforitsnaturallanguagecapabilities.Withinjusttwomonths,

itattracted100millionusers,achievingthismilestone

fasterthanTikTok(ninemonths)andInstagram(two

andahalfyears);byFebruary2025,ChatGPT’sweeklyactiveusersreached400million.4Thisunprecedentedgrowthignitedasurgeinventurecapitalinvestments

andintensifiedcompetitiontodevelopGenAIsolutionsforenhancingproductivityacrossindustries.

Inearly2025,DeepSeekemergedasamajorplayerin

theGenAIlandscape,introducingadvancedcapabilitiestoprocessandintegratemultipletypesofdata–such

asimages,sounds,andtext–simultaneously,andthe

costsandcomputingpoweraremuchlowerthanfor

earlierGenAImodels.Withinthreemonths,DeepSeekattractedover50millionusers,mirroringearlier

ChatGPTbreakthroughs.Thisinnovationreignited

investorinterestandintensifiedtheracetodevelopGenAItoolsforbroaderapplications.

Beyondtake-upbyindividualconsumers,GenAI

modelsareradicallychangingthewaybusinesses

operate.FirmsareincreasinglyleveragingGenAIfor

twopurposes:ontheproductofferingside,embeddingGenAIinproductandcustomerservicetodriveinno-vation,andontheoperationalside,GenAIredesigningtask-levelprocessesandoperationalworkflowsto

improveefficiencyandcost-effectiveness.

Businessesareincreasinglyusing

GenAItodriveinnovationandincreaseefficiency.

Figure1underscorestheaccelerationintheuseofGenAIbybusinesses.A2025globalsurveyindicatesthat71%ofrespondentshaveadoptedGenAItoolsinat

leastonebusinessfunction,risingfrom65%inearly2024and33%in2023.5

FIGURE1:AIANDGENAIUSEINBUSINESSES

OrganisationsthatuseAIinatleastonebusinessfunction,%ofrespondents*

100

80

60

40

20

0

78

72

71

56

58

50

50

65

55

47

33

20

201720182019202020212022202320242025

UseofAIUseofGenAI

*In2017,thedefinitionforAIusewasusingitinacorepartoftheorganisation?sbusinessoratscale.In2018–19,itwasembeddingatleastoneAIcapacityinbusinessprocessesorproducts.Since2020,itisthattheorganisationhasadoptedAIinatleast1function.

Source:McKinsey6

4

Hu2023

;

TechCrunch2025

.

5

McKinsey2025

.

6

Ibid

.

11

1.2Gen-AI-inducedrisks

WhileGenAIintroducesimmensebenefitsforbusi-

nesses,itscreativityandoutput-drivennatureintroducedistinctrisksthatdemandcarefulmanagement.While

someoftheserisksareamplifiedversionsofthose

fromtraditionalAI(e.g.algorithmicfairness,privacy

concerns),othersareentirelynew–particularlythoserelatedtocontentcreation,suchasalgorithmichallu-

cinations,7emergentbiases,andunauthorisedcontentreplication–andlackhistoricalparallelsinriskprofiles.8

GenAIcreatesnewriskslikethe

generationofharmfulcontentandhallucination.

RisksintroducedbyGenAIinclude,forexample,the

spreadofmisinformation(usingaudiodeepfakesto

commandsmarthomedevicesthatleadtounauthor-

isedaccess),thegenerationofharmfulcontent(with

violenceanddiscrimination),andcopyrightinfringement(usingprotectedtext,images,andmusicwithoutauthor-isationorgivingresultsthataresubstantiallysimilarin

contentandstyletoexistingworks),allofwhichposedistinctrisksforbusinessesandtheirinsurers.9

Ontheproductside,abusinessusingGenAItoolsdevelopedbytechprovidersmaysufferfinancial

harm,creatingpotentialliabilityfortheproviders.10Forexample,whenaGenAImodelintroducessecurity

vulnerabilitiesorbugsthroughitsgeneratedcode,

GenAIdevelopersfaceproductliabilityrisks.11Such

failuresinGenAIsystemscouldresemblefailuresin

criticalinfrastructure,potentiallycreatingeconomy-widesystemicrisks.AI-generatedlegalservicesmay

exposebusinessesprovidingthemtoprofessional

liabilityrisksduetoinaccurateAI-generatedcontentormisrepresentations.

Ontheoperationalside,firmsthatdeployGenAIto

steertheirbusinessesfaceriskslikeincorrect/biased

decision-making,operationalinefficiency,andfinanciallosses.

GenAIsystemsmayalsobemoresusceptibletocyber-attacks,whichcouldresultinbusinessdisruptionand

financiallosses,i.e.cybersecurityrisksstemmingfromvulnerabilitiesinGenAIsystems.

Table1showsthetypesofrisksrelatedtotraditionalandGenAI.Amongthesecategories,operational,cyberse-curity&privacy,reputational&market,andworkforce

challengesareprimarilyfirst-partyoperationalrisks,

whilebiasðicalconcerns,regulatory&compliancerisks,andESGconsiderationsalsoinvolvethird-partyproductrisk.WhilemanyrisksapplytobothtraditionalandGenAI,aspectswithspecialrelevanceforGenAIareemphasisedinthefinalcolumn.

7WhenGenAIproducesoutputsthatarefactuallyincorrect,non-sensical,orentirelydetachedfromreality,despitebeingpresentedwithhighconfidence.

8

GenRe2025

.

9

Xuetal.2024

.

10Legalliabilityforproviderstypicallyrequiresestablishingthat:1)athirdpartysufferedactualharm;2)theprovideroweda

dutyofcaretothethirdpartyandbreachedthatdutythroughnegligenceorbreachedacontractualobligation;3)thebreachwastheproximatecauseoftheharm.Importantly,manysoftwareprovidersusecontractualliabilitywaiversorlimitationsintheirtermsofservicetomitigatethisexposure.FailuresinGenAIproducts/servicescanalsocausepurelyfirst-partyopera-tionalorfinanciallossesfortheprovideritself,independentofthird-partyliability.

11Theapplicationoftraditionalproductliabilityregimestosoftware,includingAIsystems,iscomplexanduncertain.For

instance,untilrelativelyrecently,itwasunclearhowfarsoftwarecanbetreatedasaproductunderstatutessuchastheEUproductliabilitydirective.Similarly,intheUS,litigationisongoingtoestablishwhatstandardofcareattachestouseofsoftware.

12

TABLE1:AIANDGENAIRISKSFORBUSINESSES

Category

Specificrisk

TraditionalAI

GenerativeAI(Newrisksareinbold)

Operational

Algorithmicerrors;stability;reliability

Inaccuratepredictions

orunintendedoutputs

candisruptprocesses

andleadtoerrors(e.g.inventorymanagement).

GenAIoutputsmaydeviatefromintendedpurposes(offensiveorirrelevantcontent,hallucination).AsGenAIsystemsarenotvalidatedfortheirpredictivereliability,

theycausesystematicerrors,creating

heightenedriskincustomer-facingappli-cationsorautomatedcontentgeneration.Additionally,GenAIfacesheightened

servicedisruptionriskscomparedto

traditionalITsystems–itsstate-dependentworkflows(e.g.multi-turndialoguesor

contentcreation)loseprogressirreversiblyduringinterruptions.

Black-boxissuesComplexityandopacity

inAIsystemsmakeerror

tracingandaccountability

challenging,whichis

especiallyrelevantin

regulatedindustrieslike

insurance.

TraditionalAIismoreexplainablethan

GenAI.Thedecision-makingprocesses

behindGenAI’sresultsareoftendifficultor

evenimpossibletounderstand,whichmake

theprovenance,logic,andembeddedflaws

ofGenAIuntraceableandunauditableby

developers,introducingnewrisksforusers.

MaliciousattacksAIcanbeusedbythreat

attackersforinappro-priatepurposes.

GenAIcontent,suchasdeepfakesor

phishingemails,maybeexploitedfor

maliciouspurposes.ThisisdifferentfromdatapoisoningproblemswithtraditionalAI,causedbyimplantingmalicioussamples.

InadditiontoGenAIbeingusedbybad

actors,someGenAIapplicationsthem-selvesprovideabroaderattacksurface.Forexample,chatbotsthatexecute

structuredquerylanguagestatementsprovideanentrypointforattackers

throughpromptinjections.

Cybersecurity&privacy

AI-driven

cyberattacks

AIcanbeexploitedto

enhancecyberattacks,

riskingdatabreachesandoperationalsecurity.

GenAImodelsmaybemanipulatedvia

attacks(e.g.promptinjection),compro-

misingcontentqualityandsecurity(modelmanipulationrisks).

Data-privacyviolations

Collectionoflarge

amountsofdatacan

infringeonprivacylaws,leadingtolegalpenaltiesandcustomerdistrustifmishandled.

PrivacyviolationriskofGenAIishigher

thanthatoftraditionalAIasitexploresa

greatervolumeofmorecomplicatedand

unstructureddata.Theriskisparticularly

highifGenAIaccessesdataitisnot

supposedto,usesitinawayitisnot

supposedto(e.g.withoutreceivingprior

consentbytheuser),ortransfersitoutsideofthejurisdiction.GenAImayalsocauseproblemslikeprivacyintrusionthrough

constantmonitoring,heighteneddata

leakagevulnerabilities,andchallengestopersonaldatarights.

13

Category

Specificrisk

TraditionalAI

GenerativeAI(Newrisksareinbold)

Reputational&market

Customertrust&brandimage

MisuseofAIcandamagereputation,especially

ifitbreachescustomerprivacyorfairness

expectations.

Low-qualityorinaccurateGenAIoutputscanerodecustomertrustanddamage

companycredibility,asstakeholders

mayquestionthereliabilityandintentofautomatedcommunications.

Dependency&competitiverisk

Over-relianceonAIcancompromiseoperationsduringdisruptions.

GenAImayamplifytheriskasitismorecloselyandintensivelyintegratedinto

businessprocessesandmodels.

Workforce

challenges

Jobdisplacement

AIautomationmayleadtoworkforcedissatisfactionandbacklashasrolesarereplaced.

GenAIintensifiesthisrisk.AsshowninBox2,agenticAI,anadvancedversionofGenAI,willpotentiallydisplacejobsmassively.

AIskills

requirements

AIdemandsnewskills

tomanagethequality

andethicalimplicationsofoutputs,creating

challengesinworkforceupskilling.

GenAIamplifiestheproblem.Employees

mustbetrainedtointerpretGenAIoutputs,addressethicalconcerns,andensure

contentquality(‘GenAIskillrequire-ments’).GenAIalsorequirestherightbusinesscultureforadoption.

Regulatory&

compliance

EvolvingAI

regulations

Newlawsrequirebusi-

nessestoadaptquickly;non-compliancecanleadtopenalties,especially

inregulatedsectorslikeinsurance.

TheEUAIactimposescomprehensive

regulationonAIdevelopmentandusage.Regulationneedstocarefullybalance

themanagementofGenAIrisksandthepromotionoftechnologicalinnovation.

Accountability&liability

Businessesmayface

liabilityfordamage

causedbyAIsystems,

withchallengesin

assigningaccountability.

GenAImayusecopyrightedmaterial,

exposingbusinessestolegalrisksandreputationaldamage(copyrightandIP).

Biasðicalconcerns

Discrimination&bias

AIalgorithmsmay

reinforcesocietalbiases,leadingtodiscriminatorypracticesandpotential

lawsuits.

GenAIactivelycreatesnewcontent;

thus,ifusingbiaseddata,itmayproduceoutputsthatperpetuateandamplifysoci-etalstereotypes,heighteningbothethicalrisksandlitigationexposure.

Ethical

decision-making

AImayprioritiseeffi-

ciencyoverethics,

leadingtoreputational

damageifdecisionsharmcustomertrust.

GenAIoutputsmayunintentionally

violateethicalnorms,suchasgeneratingmisleadingorharmfulinformation.

ESG

Environmental&

energy

AIsystemsdemand

significantenergyand

watersupply,conflictingwithnet-zeropledges.

GenAIamplifiestheproblemasthe

underlyingtechnology,i.e.deeplearning,

demandsexponentiallygreatercomputa-

tionalpowertogeneratecomplexoutputs.Thisheightenedenergyconsumptionnot

onlystrainspowergridsbutalsointensifieswaterdependencyforcoolinghyperscaledatacentres.

Source:GenevaAssociation

14

ContentgeneratedbyGenAIsystemscomeswithacriticalelementofrandomnessandhallucinations.Thishighlightstheimportanceofmodelselection,

pre-productionevaluation,andpost-production

monitoring.Morerecently,reasoningmodels,whichgeneratenewdataratherthanjustlearningfrom

existingdata(liketraditional