Running Secure AI/ML Experiment in a Unified MLOPS-DevSecOps Pipeline
summary
Covers concrete ML‑Ops hardening: pipeline levels from training-only to end‑to‑end CI/CD, model registry and feature store additions, and a checklist of threats including data poisoning and backdoor attacks with runtime model monitoring and behavioral detection. Maps tools to controls—SageMaker vs Kubeflow for end‑to‑end workflows, Jenkins/GitLab for CI, and AWS services (VPC private endpoints, CloudWatch, GuardDuty, CloudTrail, Inspector) plus SCA and OWASP ZAP—to enforce VPC private endpoints and IAM least privilege and where to add encryption, logging, and testing.
AI-written summary · about
chapters (8)
- 0:00Introduction and speaker intro
- 1:58MLOps pipeline overview
- 5:33Integration of DevOps and MLOps
- 7:19Automation levels in MLOps
- 11:17Security risks in MLOps pipelines
- 16:23Securing data and models
- 21:46AWS security tools for MLOps
- 23:11Q&A session
related talks
50mBuilding AI Security In: MLSecOps in Practice
42mMLOps Under Attack: Threat Modeling Modern AI Systems
55mBuilding a secure MLOps pipeline, featuring PathAI
19mSecuring AI Pipelines from Development to Production
57mDevOps for Machine Learning
42mDevSecOps for AI: Introducing MLSecOps for Software 2.0
24mFrom Model to Production: Automating the ML Lifecycle with CI/CD
21mEnd-to-end secure ML Development