版权说明:本文档由用户提供并上传,收益归属内容提供方,若内容存在侵权,请进行举报或认领
文档简介
version1.1
RevisionHistory
Revision
Date
Authors
Description
.06
10/15/2024
ScottClinton,Contributors,ReviewerInputs
Re-factorSolutionsLandscapecategories,
1.0
10/15/2024
Contributors,Reviewers
FinalReleaseCandidate
1.1
12/31/2024
ScottClinton,Contributors,ReviewerInputs
NewentriesfromtheOnline
SolutionsLandscapeCatalog,Updatedlayoutand
Theinformationprovidedinthisdocumentdoesnot,andisnotintendedto,constitutelegaladvice.All
informationisforgeneralinformationalpurposesonly.Thisdocumentcontainslinkstootherthird-partywebsites.Suchlinksareonlyforconvenience,andOWASPdoesnotrecommendorendorsethecontentsofthethird-partysites.
LicenseandUsage
ThisdocumentislicensedunderCreativeCommons,CCBY-SA4.0
Youarefreeto:
●Share—copyandredistributethematerialinanymediumorformat
●Adapt—remix,transform,andbuilduponthematerialforanypurpose,evencommercially.
●Underthefollowingterms:
○Attribution—Youmustgiveappropriatecredit,providealinktothelicense,andindicateifchangesweremade.Youmaydosoinanyreasonablemannerbutnotinanywaythatsuggeststhelicensorendorsesyouoryouruse.
○AttributionGuidelines-mustincludetheprojectnameaswellasthenameoftheassetReferenced
■OWASPTop10forLLMs-LLMSecOpsSolutionsLandscape
■OWASPTop10forLLMs-CyberSecuritySolutionandLLMSecOpsLandscapeGuide
●ShareAlike—Ifyouremix,transform,orbuilduponthematerial,youmustdistributeyourcontributionsunderthesamelicenseastheoriginal.
Linktofulllicensetext:
/licenses/by-sa/4.0/legalcode
Page1
Page2
TableofContent
TableofContent2
WhoIsThisDocumentFor?
5
Objectives5
Scope
5
Introduction7
DefiningtheSecuritySolutionsLandscape8
LandscapeConsiderations8
LLMApplicationCategories,SecurityChallenges9
StaticPromptAugmentationApplications10
AgenticApplications11
LLMPlug-ins,Extensions12
ComplexApplications13
LLMDevelopmentandConsumptionModels14
LLMOpsandLLMSecOpsDefined1
5
AQuickOpsPrimer-FoundationforLLMOps15
LLMOpsLifeCycleStages-FoundationforLLMDevSecOps16
Scoping/Planning18
Page3
DataAugmentationandFine-Tuning1
9
ApplicationDevelopmentandExperimentation20
TestandEvaluation21
Release22
Deploy2
3
Operate24
Monitor25
Govern2
6
MappingtotheOWASPTop10forLLMThreatModel27
OWASPTop10forLLMsSolutionsLandscape29
EmergingGenAI/LLM-SpecificSecuritySolutions29
LLM&GenerativeAISecuritySolutions31
SolutionLandscapeMatrixDefinitions32
LandscapeSolutionMatrix33
Acknowledgements43
OWASPTop10forLLMProjectSponsors44
SilverSponsors44
References4
5
ProjectSupporters46
Page4
Letterfromtheleadauthor
Whywecreatedthiscompanionresource
ThecreationofthisdocumentwasinitiatedafterwediscussedasacoreteamthatwhiletheOWASPTop10ListforLLMsandGenerativeAIListprovidedagreatlistofrisksandpotentialmitigations,itfellshortonprovidingthenextlevelofguidance.ThisisinpartduetothestructureofwhatmakesOWASPtop10listsopopular.Thisisbeingconciseandfocusedtohighlightthetoprisksandmitigationforacertainapplicationspace.Thereweremorethanenoughcandidatestogobeyond10,butthefocusof10wefeltessentialtobeabletoensurepracticalfocusedguidance.DeviatingfromthetraditionalOWASPTop10formatwouldbloatthedocumentandimpactitsconciselisting.
Adoptingasolutionsapproachfortheproject
WhiletheTop10listforLLMandGenAIprovidesthelistTop10RiskandMitigations,wefeltitbeneficialgofurtherthantraditionalTop10ListsandtotakeasolutionsapproachandhelpconnecttheTop10RiskstotheopenssourceandcommercialsecuritysolutionsorganizationscouldlooktotohelpaddresstheTop10RisksforLLMsandGenerativeAIinapracticalway.
Inaddition,sincetheGenAIsecuritylandscapeismovingsoquickly,coveringarangeofnewapplication
typesfromstaticpromptaugmentation,throughRAG,pluginsandAgenticAIarchitectures,wesawarangeofnewsecuritysolutionsemergingandwantedtobeabletoprovidearegularlyupdatedresourcetoidentifythesolutionthatcouldbeusedtoaddressthesenewarchitecturesandapplicationriskshighlightedintheTop10forLLMandGenAIList.
Structuringthedocument
Toorganizethesolutions,wechosetoleverageanddocumenttheapplicationtypesandtheLLM/GenAIOpsandSecOpslifecycleandcategoriestoprovideanactionablewaytobothorganizethesolutionsandmap
themtotheTop10forLLMandGenAI,whichwewouldupdatequarterly.Toaccompanythisdocumentwealsodecidedtopublishan
onlinedirectory
.WehopethissolutionguideishelpfulinimplementingyourownstrategyforsecureLLMandGenAIadoptionwithinyourorganization.
-ScottClinton
Co-LeadOWASPTop10forLLMProject
&AI,SecuritySolutionsInitiativeLead
Page5
WhoIsThisDocumentFor?
Thisdocumentistailoredforadiverseaudiencecomprisingdevelopers,AppSecprofessionals,DevSecOpsandMLSecOpsteams,dataengineers,datascientists,CISOs,andsecurityleaderswhoarefocusedon
developingstrategiestosecureLargeLanguageModels(LLMs)andGenerativeAIapplications.ItprovidesareferenceguideofthesolutionsavailabletoaidinsecuringLLMapplications,equippingthemwiththe
knowledgeandtoolsnecessarytobuildrobust,secureAIapplications.
Objectives
ThisdocumentisintendedtobeacompaniontotheOWASPTop10forLargeLanguageModel(LLM)
ApplicationsListandtheCISOCybersecurity&GovernanceChecklist.Itsprimaryobjectiveistoprovidea
referenceresourcefororganizationsseekingtoaddresstheidentifiedrisksandenhancetheirsecurity
programs.Whilenotdesignedtobeanall-inclusiveresource,thisdocumentoffersaresearchedpointof
viewbasedonthetopsecuritycategoriesandemergingthreatareas.Itcapturesthemostimpactfulexistingandemergingcategories.Bycategorizing,defining,andaligningapplicabletechnologysolutionareaswiththeemergingLLMandgenerativeAIthreatlandscape,thisdocumentaimstosimplifyresearcheffortsandserveasasolutionsreferenceguide.
Scope
Thescopeofthisdocumentistocreateashareddefinitionofsolutioncategoryareasthataddressthe
securityoftheLLMandgenerativeAIlifecycle,fromdevelopmenttodeploymentandusage.ThisalignmentsupportstheOWASPTop10ListForLLMsoutcomesandtheCISOCybersecurityandGovernanceChecklist.Toachievethis,thedocumentwillcreateaninitialframeworkandcategorydescriptors,utilizingboth
open-sourcesolutionsandprovidingmechanismsforsolutionproviderstoaligntheirofferingswithspecificcoverageareasasexamplestosupporteachcategory.
Page6
Thedocumentadherestoseveralkeyrulestomaintainitsintegrityandusefulness:
●Vendor-AgnosticandOpenApproach:Itmaintainsaneutralstance,avoidingrecommendationsofonetechnologyoveranother,insteadprovidingcategoryguidancewithchoicesandoptions.
●Straightforward,ActionableGuidance:Thedocumentoffersclear,actionableadvicethatorganizationscanreadilyimplement.
●CoordinatedKnowledgeGraph:Itincludescoordinatedterms,definitions,anddescriptionsforkeyconcepts.
●PointtoExistingStandards:Whereexistingstandardsorsourcesoftruthareavailable,the
documentreferencestheseinsteadofcreatingnewsources,ensuringconsistencyandreliability.
Page7
Introduction
WiththegrowthofGenerativeAIadoption,usage,andapplicationdevelopmentcomesnewrisksthataffecthoworganizationsstrategizeandinvest.Astheserisksevolve,sodoriskmitigationsolutions,technologies,frameworks,andtaxonomies.Toaidsecurityleadersinprioritization,conversationsaboutemerging
technologyandsolutionareasmustbealignedappropriatelytoclearlyunderstoodbusinessoutcomesforAIsecuritysolutions.ThebusinessoutcomesofAIsecuritysolutionsmustbeproperlydefinedtoaidsecurityleadersinbudgeting
Manyorganizationshavealreadyinvestedheavilyinvarioussecuritytools,suchasvulnerabilitymanagementsystems,identityandaccessmanagement(IAM)solutions,endpointsecurity,DynamicApplicationSecurityTesting(DAST),observabilityplatforms,andsecureCI/CD(ContinuousIntegration/ContinuousDeployment)tools,tonameafew.However,thesetraditionalsecuritytoolsmaynotbesufficienttofullyaddressthe
complexitiesofAIapplications,leadingtogapsinprotectionthatmaliciousactorscanexploit.Forexample,traditionalsecuritytoolsmaynotsufficientlyaddresstheuniquedatasecurityandsensitiveinformation
disclosureprotectioninthecontextofLLMandGenAIapplications.Thisincludesbutisnotlimitedtothechallengesofsecuringsensitivedatawithinprompts,outputs,andmodeltrainingdata,andthespecificmitigationstrategiessuchasencryption,redaction,andaccesscontrolmechanisms.
EmergentsolutionslikeLLMFirewalls,AI-specificthreatdetectionsystems,securemodeldeployment
platforms,andAIgovernanceframeworksattempttoaddresstheuniquesecurityneedsofAI/ML
applications.However,therapidevolutionofAI/MLtechnologyanditsapplicationshasdrivenanexplosionofsolutionapproaches,whichhasonlyaddedtotheconfusionfacedbyorganizationsindeterminingwheretoallocatetheirsecuritybudgets.
Page8
DefiningtheSecuritySolutionsLandscape
TherehavebeenmanyapproachestocharacterizingthesolutionslandscapeforLargeLanguageModeltoolsandinfrastructure.InordertodevelopasolutionslandscapethatfocusesonthesecurityofLLMapplicationsacrossthelifecyclefromplanning,development,deployment,andoperation,therearefourkeyareasofinputwehavefocusedontodevelopbothadefinitionforLargeLanguageModelDevSecOPsandrelatedsolutionslandscapecategories.
LandscapeConsiderations
ApplicationTypesandScope-whichimpactsthepeople,processes,andtoolsneededbasedonthecomplexityoftheapplicationandtheLLMenvironment,as-a-service,self-hosted,orcustom-built.
EmergingLLMSecOpsProcess-whilethisisaworkinprogress,manyarelookingtoadaptandadoptexistingDevOpsandMLOpsandassociatedsecuritypractices.WeexpectourdefinitiontoevolveasthedevelopmentprocessesforLLMapplicationsbegintomature.
ThreatandRiskModeling-understandingtherisksposedbyLLMsystems,applicationusage,ormisuselikethoseoutlinedintheOWASPTop10forLLMsandGenerativeAIApplications,arekeytounderstandingwhichsolutionsarebestsuitedtoimprovethesecuritypostureandcombatarangeofattacks.
TrackingEmergingSolutions-manyexistingsecuritysolutionsareadaptingtosupportLLMdevelopmentworkflowsandusecaseshowevergiventhenatureofnewthreatsandevolvingtechnologyandarchitecturesnewtypesofLLM-specificsecuritysolutionswillbenecessary.
Page9
LLMApplicationCategories,SecurityChallenges
OrganizationshavebeenleveragingMachineLearninginapplicationsfordecades.Thisoftenrequired
detailedexpertiseinDataScienceandextensivemodeltraining.GenerativeAIhaschangedthis.Specifically,LargeLanguageModels(LLMs)havemademachinelearningtechnologywidelyaccessible.Theabilityto
dynamicallyinteractinplainlanguagehasopenedthedoorforthecreationofanewclassofdata-driven
applicationsandapplicationintegrations.Furthermore,usageisnolongerlimitedtothehighlyskilledeffortsoftraditionaldevelopersanddatascientists.Pre-trainedmodelsenablenearlyanyonetoperformcomplexcomputationaltasks,regardlessofpriorexposuretoprogrammingorsecurity.Organizationshavebeen
leveragingMachineLearninginapplicationsfordecadesincludingNaturalLanguageProcessing(NLP)modelsthatoftenrequiredetailedexpertiseinDataScienceandextensivemodeltraining.
Withtheadventoftransformerstechnologyenablinggenerativecapabilitiescombinedwiththeeaseofaccessforpre-trainedas-a-servicemodelslikeChatGPTandotheras-a-service,FourmajorcategoriesofLLMApplicationArchitectureemerged;Prompt-centric,AIAgents,Plug-ins/extensions,andcomplex
generativeAIapplicationwheretheLLMplaysakeyroleinalargerapplicationusecase.
(figure:ApplicationCategories&SummaryAttributes)
HavingacommonviewoftypicalLLMapplicationarchitectures,includingagents,models,LLMs,andtheMLapplicationstack,iscrucialfordefiningandaligningtheapplicationstack,securitymodel,andapplicationofferings.Below,wehaveprovidedashortdescriptionofkeycharacteristics,usecases,andsecurity
challengesforeachapplicationcategory.
StaticPromptAugmentationApplications
Theseapplicationsinvolvespecificstaticnaturallanguageinputstoguidethebehaviorofalargelanguagemodel(LLM)towardgeneratingthedesiredoutput.Thistechniqueoptimizestheinteractionbetweenthe
userandthemodelbyfine-tuningthephrasing,context,andinstructionsgiventotheLLM.These
applicationsallowuserstoaccomplishawiderangeoftasksbysimplyrefininghowtheyaskquestionsorprovideinstructions.
KeyCharacteristics
●Humantomodel/modeltohumaninteractionandresponse
●Staticpromptaugmentation
●FlexibilityandCreativity
●SimplicityandAccessibility
●RapidPrototypingandExperimentation
UseCaseExamples
●Experimentation/RapidPrototyping
●ContentGenerationTools
●TextSummarizationApplications
●Question-AnsweringSystems
●LanguageTranslationTools
●ChatbotsandVirtualAssistants
SecurityChallenges
●Prompt-basedapplicationsfacesecurityriskslikepromptinjectionattacksanddataleakagefrompoorlycraftedprompts.Lackofcontextorstatemanagementcanleadtounintendedoutputs,
increasingmisusevulnerability.User-generatedpromptsmaycauseinconsistentorbiasedresponses,riskingcomplianceorethicalviolations.Ensuringpromptintegrity,robustinputvalidation,andsecuringtheLLMenvironmentarecrucialtomitigatetheserisks.
Page10
Page11
AgenticApplications
TheseapplicationsleverageLargeLanguageModels(LLMs)toautonomouslyorsemi-autonomouslyperformtasks,makedecisions,andinteractwithusersorothersystems.Theseagentsaredesignedtoactonbehalfofusers,handlingcomplexprocessesthatofteninvolvemultiplesteps,integrations,andreal-time
decision-making.Theyoperatewithalevelofautonomy,allowingthemtocompletetaskswithoutconstanthumanintervention.
KeyCharacteristics
●AutonomyandDecision-Making
●InteractionwithExternalSystems
●StateManagementandMemory
●ComplexWorkflowAutomation
●Human-AgentCollaboration
UseCaseExamples
●VirtualAssistants
●CustomerSupportBots
●ProcessAutomationAgents
●DataAnalysisandReportingAgents
●IntelligentPersonalizationAgents
●SecurityandComplianceAgents
SecurityChallenges
●Agentapplications,withtheirautonomyandaccesstovarioussystems,mustbecarefullysecuredtopreventmisuse.Theyfacesecuritychallengeslikeunauthorizedaccess,increasedexploitationrisksduetointeractionwithmultiplesystems,andvulnerabilitiesindecision-makingprocesses.If
someonegainscontrolofanautonomousagent,theconsequencescouldbesevere,especiallyin
criticalsystems.Ensuringrobustaccesscontrolsandencryptionmethodstoprotectagainstthisisessential.Ensuringdataintegrityandconfidentialityiscritical,asagentsoftenhandlesensitive
informationitisimportanttosecuredataatallstages,includingat-rest,inmotion,andaccess
throughsecuredAPIs.Theirautonomyalsoposesrisksofunintendedorharmfuldecisionswithoutoversight.Robustauthentication,encryption,monitoring,andfail-safemechanismsareessentialtomitigatethesesecurityrisks.ObservabilityandTraceabilitysolutionsthatmonitortheentire
lifecycleoftheAgents(Design,Development,Deployment,andVisibilityondecision-making)mustbeconsideredtoensurereal-timecorrectionsusingahumans-in-the-loopprocesscanbeenforced.
Page12
LLMPlug-ins,Extensions
Plug-insareextensionsoradd-onsthatintegrateLLMsintoexistingapplicationsorplatforms,enablingthem
toprovideenhancedornewfunctionalities.Plug-instypicallyserveasabridgebetweentheLLMandtheapplication,facilitatingseamlessintegration,suchasaddingalanguagemodeltoawordprocessorfor
grammarcorrectionorintegratingwithcustomerrelationshipmanagement(CRM)systemsforautomatedemailresponses.
WhileitcanbesometimesdifficulttodrawthelinebetweenAgentsandplug-insorextensionswhichareoftencomponentsoflargerapplications,onemeasureisthewayitisdeployedandused.Forexample,a
plug-inwouldbeapre-builtagentdesignedforreusethatyoucallexplicitly,throughanAPI,oraspartofanLLMspluginorextensionframeworkvs.customcoderunninginthebackgroundonaperiodicbasis.
KeyCharacteristics
●ModularityandFlexibility
●SeamlessIntegration
●TaskSpecificFocus
●EaseofDeploymentandUse
●RapidUpdatesandMaintenance
UseCaseExamples
●ContentGenerationTools
●TextSummarizationApplications
SecurityChallenges
●Pluginsinteractingwithsensitivedataorcriticalsystemsmustbecarefullyvettedforsecurity
vulnerabilities.Poorlydesignedormaliciouspluginscancausedatabreachesorunauthorized
access.LLMpluginsfacechallengeslikecompatibilityissues,whereupdatescanintroduce
vulnerabilities,andintegrationwithsensitivesystemsincreasestheriskofdataleaks.Ensuring
secureAPIinteractions,regularupdates,androbustaccesscontrolsiscrucial.Resource-intensivepluginsmaydegradeperformance,riskingexploitation.
Page13
ComplexApplications
ComplexapplicationsaresophisticatedsoftwaresystemsthatdeeplyintegrateLargeLanguageModels
(LLMs)asacentralcomponenttoprovideadvancedfunctionalitiesandsolutions.Theseapplicationsare
characterizedbytheircomprehensivescope,scalability,andtheintegrationofmultipletechnologiesandcomponents.Theyaretypicallydesignedtosolveintricateproblems,ofteninenterpriseenvironments,andrequireextensivedevelopment,engineering,andongoingmaintenanceefforts.
KeyCharacteristics
●Multi-componentarchitecturesaredesignedtoprocesspromptsfromothernon-humansystems.
●Oftenusemultipleintegrations,includingothermodels.
●Multi-ComponentArchitecture
●ScalabilityandPerformance
●AdvancedFeaturesandCustomization
●End-to-EndWorkflowAutomation
UseCaseExamples
●LegalDocumentAnalysisPlatforms
●AutomatedFinancialReportingSystems
●CustomerServicePlatforms
●HealthcareDiagnostics
SecurityChallenges
●ComplexLLMapplicationsfacemajorsecuritychallengesduetotheirintegrationwithmultiple
systemsandextensivedatahandling.TheseincludeAPIvulnerabilities,databreaches,and
adversarialattacks.Thecomplexityincreasestheriskofmisconfigurations,leadingtounauthorizedaccessordataleaks.Managingcomplianceacrosscomponentsisalsodifficult.Robustencryption,accesscontrols,regularsecurityaudits,andcomprehensivemonitoringareessentialtoprotect
theseapplicationsfromsophisticatedthreatsandensuredatasecurity.
Page14
LLMDevelopmentandConsumptionModels
OneofthefirstconsiderationsforanorganizationisdecidingupontheapproachtoleveragingLLM
capabilitiesbasedonthetypeofapplicationandgoalsfortheproject.Today,developershaveachoiceoftwoprimarydeploymentmodelswhenimplementingLLM-basedapplicationsandsystems.
CreateaNewModel:ThetrainingprocessforcustomLLMsisintensive,ofteninvolvingdomain-specificdatasetsandextensivefine-tuningtoachievedesiredperformancelevels.ThisapproachismoreakintoMLOpsbuildingMLmodelsfromthegroundup,withdetaileddataanalysis,collectionformatting,cleaning,andlabeling.Oneofthebenefitsofthisapproachisthatyouknowthelineageandsourceofthedatathemodelisbuiltonandcanattestdirectlytoitsvalidityandfit.However,amajordownsideistheresources,cost,andexpertisenecessarytobuild,train,andverifyamodelthatmeetstheprojectobjectives.CustomLLMsprovidetailoredsolutionsoptimizedforspecifictasksanddomains,offeringhigheraccuracyand
alignmentwithanorganization'sspecificneeds.
ConsumeandCustomizeExistingModels:Pre-trained(foundation)models,whetherself-hostedorofferedasaservice,suchaswithChatGPT,Bertandothersontheotherhandprovideamoreaccessibleentrypointfororganizations.ThesemodelscanbequicklydeployedviaAPIs,allowingforrapidsolutionvalidationand
integrationintoexistingsystems.TheLLMOpsprocessinthisscenarioemphasizescustomizationthrough
fine-tuningwithspecificdatasets,ensuringthemodelmeetstheapplication'suniquerequirements,followedbyrobustdeploymentandmonitoringtomaintainperformanceandsecurity.
Page15
LLMOpsandLLMSecOpsDefined
HavingacommonviewoftypicalLLMapplicationarchitectures,includingagents,models,LLMs,andtheMLapplicationstack,iscrucialfordefiningandaligningtheapplicationstackandsecuritymodel.
(figure:LLMOpsrelatedOperationsProcessforData,MachineLearningandDevOps)
AQuickOpsPrimer-FoundationforLLMOps
DevOps,whichemphasizescollaboration,automation,andcontinuousintegrationanddeployment(CI/CD),haslaidthegroundworkforefficientsoftwaredevelopmentandoperations.Bystreamliningthesoftwaredevelopmentlifecycle,DevOpsenablesrapidandreliabledeliveryofapplications,fosteringacultureof
collaborationbetweendevelopmentandoperationsteams.
DataOpsbuildsonDevOps,wheredatapipelinesaremanagedwithsimilarautomation,versioncontrol,andcontinuousmonitoring,ensuringdataqualityandcomplianceacrossthedatalifecycle.MLOpsalsoextendstheDevOpsprinciplestomachinelearning,focusingontheuniquechallengesofmodeldevelopment,
training,deployment,andmonitoring.UtilizingDevOpsasafoundationensuresthatbothDataOpsandMLOpsinheritarobustinfrastructurethatprioritizesefficiency,scalability,security,andfasterinnovationin
data-drivenandmachinelearningapplications.
MLOpsandDataOpsarefoundationaltoLLMOpsbecausetheyestablishthecriticalprocessesand
infrastructureneededformanagingthelifecycleoflargelanguagemodels(LLMs).DataOpsensuresthatdatapipelinesareefficientlymanaged,fromdatacollectionandpreparationtostorageandretrieval,providing
high-quality,consistent,andsecuredatathatLLMsrelyonfortrainingandinference.MLOpsextendsthese
Page16
principlesbyautomatingandorchestratingthemachinelearninglifecycle,includingmodeldevelopment,training,deployment,andmonitoring.
LLMOpsandMLOps,whilerootedinthesamefoundationalprinciplesoflifecyclemanagement,diverge
significantlyintheirfocusandrequirementsduetothespecificdemandsoflargelanguagemodels(LLMs).
LLMOpsencompassesthecomplexitiesoftraining,deploying,andmanagingLLMs,whichrequiresubstantialcomputationalresourcesandsophisticatedhandling.LLMOpsensurethatLLMsareefficientlyintegrated
intoproductionenvironments,monitoredforperformanceandbiases,andupdatedasneededtomaintain
theireffectiveness.ThisholisticapproachensuresthatthedeploymentandoperationofLLMsare
streamlined,scalable,andsecure,includingconsiderationsfordatavalidationandprovenancetoensurethatthedatausedfortrainingandfine-tuningLLMsistrustworthyandfreefromtampering.Thiscaninclude
techniquesfordataauditingandverification.
LLMOpsLifeCycleStages-Foundation
温馨提示
- 1. 本站所有资源如无特殊说明,都需要本地电脑安装OFFICE2007和PDF阅读器。图纸软件为CAD,CAXA,PROE,UG,SolidWorks等.压缩文件请下载最新的WinRAR软件解压。
- 2. 本站的文档不包含任何第三方提供的附件图纸等,如果需要附件,请联系上传者。文件的所有权益归上传用户所有。
- 3. 本站RAR压缩包中若带图纸,网页内容里面会有图纸预览,若没有图纸预览就没有图纸。
- 4. 未经权益所有人同意不得将文件中的内容挪作商业或盈利用途。
- 5. 人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对用户上传分享的文档内容本身不做任何修改或编辑,并不能对任何下载内容负责。
- 6. 下载文件中如有侵权或不适当内容,请与我们联系,我们立即纠正。
- 7. 本站不保证下载资源的准确性、安全性和完整性, 同时也不承担用户因使用这些下载资源对自己和他人造成任何形式的伤害或损失。
评论
0/150
提交评论