cyber.hendryadrian.com
Security Report · wiz.io

Wiz Report Finds Rapid Growth in Cloud AI Adoption and Persistent Security Risks

· Original article ↗

Summary

Wiz analyzed cloud account data and found rising adoption of AI services and self-hosted models. The report also highlights security vulnerabilities and a DeepSeek database exposure, and calls for better visibility and controls.

Key points

  • Wiz says 74% of organizations use managed AI services and 85% host some AI technology.
  • Self-hosted AI model adoption rose from 42% to 75% over the past year, according to the report.
  • DeepSeek model adoption tripled in under two weeks after DeepSeek-R1’s release; 7% of organizations use its self-hosted models.
  • Wiz reports that an exposed DeepSeek database leaked sensitive chat history and log streams.
  • The report highlights vulnerabilities involving NVIDIA AI, SAP AI Core, Ollama, and services from Hugging Face, Replicate, and SAP.
  • Wiz recommends improving visibility into AI deployments and using controls to reduce data exposure and unauthorized use.

Article Details

Publisher
Wiz Research
Report Period
2025
Scope
Cloud AI adoption and security risks
Sample Size
The article reports analysis of over 150,000 cloud accounts, and later describes the sample as hundreds of thousands of public cloud accounts.
Key Statistics
  • 74% of organizations use managed AI services, up from 70% the previous year.
  • 85% of organizations host some form of AI technology.
  • Self-hosted AI model adoption rose from 42% to 75% over the past year.
  • DeepSeek self-hosted AI models are used by 7% of organizations; adoption reportedly tripled in under two weeks after DeepSeek-R1 was released.
  • OpenAI or Azure OpenAI SDKs are used in 63% of cloud environments, up from 53% the previous year; 8 of the 10 most popular hosted AI technologies are linked to open-source projects.
Recommendations
  • Improve visibility into AI deployments to manage risks such as Shadow AI.
  • Establish controls to prevent data exposure and unauthorized use.

CVE

CVE-2024-0132CVE-2024-0132 – A critical NVIDIA AI vulnerability affecting over 35% of cloud environments.

Vendors

Hugging FaceAI PaaS vulnerabilities – Flaws in services provided by Hugging Face, Replicate, and SAP could have allowed attackers to access customer data.NvidiaCVE-2024-0132 – A critical NVIDIA AI vulnerability affecting over 35% of cloud environments.OpenAIOpenAI Still Leads, But Open-Source AI Gains Traction ReplicateAI PaaS vulnerabilities – Flaws in services provided by Hugging Face, Replicate, and SAP could have allowed attackers to access customer data.SAPSAPwned – A flaw in SAP AI Core that could allow attackers to take over services and access customer data.WizIn our second annual State of AI in the Cloud report, the Wiz Research team analyzed aggregated data from over 150,000 cloud accounts to explore the evolving AI landscape.

Products

Azure OpenAI SDKsOpenAI remains the most widely used AI provider, with 63% of cloud environments leveraging OpenAI or Azure OpenAI SDKs, up from 53% last year. However, open-source AI continues to play a pivotal role, with 8 of the 10DeepSeekThis year’s findings highlight the rapid growth of self-hosted AI models, the rise of DeepSeek, and the persistent need for stronger security measures.DeepSeek-R1meteoric rise of DeepSeek. Adoption of its models tripled in under two weeks following the release of DeepSeek-R1, with 7% of organizations now using its self-hosted AI models. DeepSeek’s rapid ascent highlights theOllamaProbllama – A remote code execution vulnerability in the open-source AI infrastructure project Ollama.SAP AI CoreSAPwned – A flaw in SAP AI Core that could allow attackers to take over services and access customer data.Wiz’s AI Security Posture Management (AI-SPM) solutionWiz’s AI Security Posture Management (AI-SPM) solution helps teams gain full visibility into AI adoption, mitigate risks, and enable secure innovation. As AI continues to reshape the cloud landscape, proactive security

Related Articles