AI代理的兴起：评估与治理的行动框架.pptxVIP

Incollaboration

withCapgemini

AIAgentsinAction:

FoundationsforEvaluationandGovernance

WHITEPAPER

NOVEMBER2025

Images:AdobeStock

Contents

Foreword4

Executivesummary5

Introduction6

1EvolvingtechnicalfoundationsofAIagents8

1.1ThesoftwarearchitectureofanAIagent8

1.2Communicationprotocolsandinteroperability10

1.3Cybersecurityconsiderations12

2FoundationsforAIagentevaluationandgovernance13

2.1Classification14

2.2Evaluation19

2.3Riskassessment22

2.4GovernanceconsiderationsforAIagents:aprogressiveapproach25

3Lookingahead:multi-agentecosystems29

Conclusion30

Contributors31

Endnotes34

Disclaimer

Thisdocumentispublishedbythe

WorldEconomicForumasacontribution

toaproject,insightareaorinteraction.

Thefindings,interpretationsand

conclusionsexpressedhereinarearesult

ofacollaborativeprocessfacilitatedand

endorsedbytheWorldEconomicForum

butwhoseresultsdonotnecessarily

representtheviewsoftheWorldEconomic

Forum,northeentiretyofitsMembers,

Partnersorotherstakeholders.

©2025WorldEconomicForum.Allrights

reserved.Nopartofthispublicationmay

bereproducedortransmittedinanyform

orbyanymeans,includingphotocopying

andrecording,orbyanyinformation

storageandretrievalsystem.

AIAgentsinAction:FoundationsforEvaluationandGovernance2

Inrecentyears,organizationshavemovedbeyondpredictivemodelsandchatinterfacestoexperimentwithartificialintelligence(AI)inmoretransformativeways.AIagentsarenowemergingasintegrated

collaboratorsinbusiness,publicservicesand

everydaylife.Theadoptionof