OpenAI：2025AI模型恶意使用的对抗策略与案例分析报告 英文版 .docxVIP

February2025

Disruptingmalicioususesofourmodels:anupdateFebruary2025

Tableofcontents

ExecutiveSummary 3

TheuniquevantagepointofAIcompanies 5

Sharingasaforcemultiplier 6

Casestudies 7

Surveillance:“PeerReview” 7

DeceptiveEmploymentScheme 10

In?uenceactivity:“SponsoredDiscontent” 13

Romance-baitingscam(“pigbutchering”) 18

Iranianin?uencenexus 23

Cyberthreatactors 27

Covertin?uenceoperation 29

Taskscam 32

Authors 35

Disruptingmalicioususesofourmodels:anupdateFebruary2025 2

ExecutiveSummary

Ourmissionistoensurethatarti?cialgeneralintelligencebene?tsallofhumanity.We

advancethismissionbydeployingourinnovationstobuildAItoolsthathelppeoplesolve

reallyhardproblems.

AswelaidoutinourEconomicBlueprintinJanuary,webelievethatmakingsureAIbene?ts

themostpeoplepossiblemeansenablingAIthroughcommon-senserulesaimedat

protectingpeoplefromactualharms,andbuildingdemocraticAI.Thisincludespreventing

useofAItoolsbyauthoritarianregimestoamasspowerandcontroltheircitizens,orto

threatenorcoerceotherstates;aswellasactivitiessuchascovertin?uenceoperations(IOs),

childexploitation,scams,spam,andmaliciouscyberactivity.TheAI-poweredinvestigative

capabilitiesthat?owfromOpenAI’sinnovationsprovidevaluabletoolstohelpprotect

democraticAIagainstthemeasuresofadversarialauthoritarianregimes.

IthasnowbeenayearsinceOpenAIbecamethe?rstAIresearchlabtopublishreportson

ourdisruptionsinane?orttosupportbroadere?ortsbyU.S.andalliedgovernments,industry

partners,andotherstakeholders,topreventabusebyadversariesandothermaliciousactors.

ThislatestreportoutlinessomeofthetrendsandfeaturesofourAI-poweredwork,together

withcasestudiesthathighlightthetypesofthreatswe’vedisrupted.

OurFebruaryreportincludestwodisruptionsinvolvingthreatactorsthatappeartooriginate

fromChina(se