HM政府：到2025年生成人工智能的安全和安保风险（英文版）（6页）.pdf

1、1Safety and Security Risks of Generative Artificial Intelligence to 20252SummaryGenerative AI development has the potential to bring significant global benefits.But it will also increase risks to safety and security by enhancing threat actor capabilities and increasing the effectiveness of attacks.T

2、hedevelopmentandadoptionofgenerativeAItechnologieshasthepotentialtobringsubstantialbenefitsifmanagedappropriately.Productivityandinnovationacrossmanysectorsincludinghealthcare,financeandinformationtechnologywillaccelerate.GenerativeAIwillalsosignificantlyincreaseriskstosafetyandsecurity.By2025,gener

3、ativeAIismorelikelytoamplifyexistingrisksthancreatewhollynewones,butitwillincreasesharplythespeedandscaleofsomethreats.Thedifficultyofpredictingtechnologicaladvancescreatessignificantpotentialfortechnologicalsurprise;additionalthreatswillalmostcertainlyemergethathavenotbeenanticipated.Therapidprolif

4、erationandincreasingaccessibilityofthesetechnologieswillalmostcertainlyenableless-sophisticatedthreatactorstoconductpreviouslyunattainableattacks.Risksinthedigitalsphere(e.g.cyber-attacks,fraud,scams,impersonation,childsexualabuseimages)aremostlikelytomanifestandtohavethehighestimpactto2025.Riskstop

5、oliticalsystemsandsocietieswillincreaseinlikelihoodasthetechnologydevelopsandadoptionwidens.Proliferationofsyntheticmediariskserodingdemocraticengagementandpublictrustintheinstitutionsofgovernment.PhysicalsecurityriskswilllikelyriseasGenerativeAIbecomesembeddedinmorephysicalsystems,includingcritical

6、infrastructure.Theaggregateriskissignificant.Thepreparednessofcountries,industriesandsocietytomitigatetheserisksvaries.Globallyregulationisincompleteandhighlylikelyfailingtoanticipatefuturedevelopments.Our definitions and scopeSafety and Security:Theprotection,wellbeingandautonomyofcivilsocietyandth