2025年第二季度AI采用与风险报告

Table of contentsAI ADOPTION AND RISK REPORTKey findingsALADOPTIONAND RISK REPORTTable of contents Growth in AI usageThe top 25 AI tools in the workplaceGrowth in Al usage AI adoption by industryRisk assessment of AI toolsThe top 10 AI tools by risk levelDeepSeek usage by end usersWhat corporate data workers put into AI toolsHow AI-generated content is used at workHow employees of different seniority use AI at workHow developers use AI to write codeDeepSeek model usage in development projectsAl adoption by industryRisk assessment of Al toolsThe top 10 Al tools by risk levelDeepSeek usage by end usersWhat corporate data workers put into Al toolsHow Al-generated content is used at workHow employees of different seniority use Al at workHow developers use Al to write codeDeepSeek model usage in development projects IntroductionSince ChatGPT launched in volume and sensitivity of data flowing into these systems has grownexponentially. Companies now face a dual challenge: harnessing AI'spotential while managing the substantial data risks it introduces. The majority of current AI usage falls under what’s called "shadow AI"– the use of AI tools unsanctioned by corporate IT departments. Thisgrassroots adoption is predominantly driven by younger, mid-levelemployees and individual contributors, while adoption continues to 2022, AI adoption has becomeone of the fastest growingworkplace technologies inhistory. What began as individualemployees experimenting withgenerative AI has shifted tothese tools becoming embeddedin core business processesacross organizations of all sizes.The speed of this transition fromnovelty to necessity isunprecedented.lag among more senior employees who tend to more cautious aboutnew technologies. This comprehensive analysis fromCyberhaven Labs draws on actual AIusage patterns of 7 million workers,providing an unprecedented view into theadoption patterns and securityimplications of AI in the corporate For forward-thinking organizations, a significant opportunity exists inunderstanding and leveraging this grassroots AI usage. By identifyinghow employees are successfully using AI to increase efficiency anddrive business results, companies can strategically implement thesetools and methodologies on a broader scale, capturing their benefitsenvironment. enable AI adoption, they must also implement robust guardrails toprotect their most sensitive information assets. 01AI usage in the workplaceKey FindingsKey Findings 71.7% of AI tools are high or critical risk, with 39.5% of AItools inadvertently exposing user interaction/training dataand 34.4% exposing user data. 71.7% of Al tools are high or critical risk, with 39.5% of Altools inadvertently exposing user interaction/training data instead of enterprise-ready tools (low and very low risk).DeepSeek usage surged andthen plateaued End user use of DeepSeek through deepseek.com surged inthe 3 weeks after the R1 release, but soon plateaued endingthe first 7 weeks at 672.8% growth. instead of enterprise-ready tools (low and very low risk).DEEPSEEK USAGE SURGED ANDEnd user use of DeepSeek through deepseek.com surged inTHEN PLATEAUEDthe 3 weeks after the R1 release, but soon plateaued ending Use of DeepSeek in local application development by AIengineers peaked at 17.7% of AI developer usage in FebruaryUse of DeepSeek in local application development by Al 34.8% of corporate data employees put into AI tools issensitive, up from 27.4% a year ago and 10.7% two years ago. The most common types of sensitive data employees put34.8% of corporate data employees put into Al tools issensitive, up from 27.4% a year ago and 10.7% two years ago. materials (17.1%) and Sales andMarketing data (10.7%).When companies roll out AI developer tools likeCursor ormaterials (17.1%) and Sales and Marketing data (10.7%) At the same time, when companies roll out AI developerAt the same time, when companies roll out Al developer Mid-level employees like analysts and specialists use AItools 3.5 times as much as the next-nearest cohortMid-level employees like analysts and specialists use Altools 3.5 times as much as the next-nearest cohort Among software engineers, the highest AI usage is amongmid-level software engineers, who use AI 189% more thantheir morejunior counterparts. Among software engineers, the highest Al usage is amongmid-level software engineers, who use Al 189% more thantheir more junior counterparts. 1 Growth in AI usage fastest rates of adoption for any new technology in the workplace,substantially outpacing even SaaS adoption, which took years toachieve similar penetration levels. Al usage at work is growing exponentially(Data ingress and egress events per 100,000 employees) The top 25 AI tools in the workplace2024 Ranking2025 RankingCyberhaven continuously tracks usage metricsThe top 25 Al tools in the workplaceCyberhaven continuously tracks usage metrics2024 Ranking2025 Ranking ¡¬ ChatGP¥¢¬ BarŸ¨¬ Gemin—¬ Po¹š¬ Perplexity A§significant shift