ML Security Operations at One of the Largest Brewing Companies
summary
Heineken runs a secure ML workbench that locks down internet package installs by serving vetted packages from an internal index and quarantining suspicious uploads with Sonatype Nexus + Nexus Firewall. Pipelines live in Azure ML Studio with code in Azure DevOps, secrets in Key Vault, logging to Application Insights/New Relic, and models tracked via CI/CD across DTAP environments. A global analytics team of ~100 scales these pipelines to 100+ countries, claiming automated policy enforcement stops vulnerable PyPI packages before deployment.
AI-written summary · about
chapters (14)
- 0:00Intro
- 4:42Data democratization
- 5:20MLOps: Machine learning operations
- 6:03Machine learning operations life cycle
- 9:14Shift-left security
- 9:56What is MLSecOps?
- 14:39Use case: Secure ML workbench
- 15:34Use case: Insecure ML workbench
- 18:49Quiz
- 19:55Security importance
- 21:10Sonatype Nexus
- 22:39Use case: Secure ML workbench
- 23:50Conclusion
- 24:42Outro
related talks
43mHacking and Securing ML Systems and Environments
26mRunning Secure AI/ML Experiment in a Unified MLOPS-DevSecOps Pipeline
50mBuilding AI Security In: MLSecOps in Practice
19mSecuring AI Pipelines from Development to Production
55mBuilding a secure MLOps pipeline, featuring PathAI
42mMLOps Under Attack: Threat Modeling Modern AI Systems
20mNo API Left Untested: Shift Left with API Security
15mML anywhere: Single pane of glass for cross-Region & on-premises AI/ML (PRT066)