Secure and compliant machine learning for regulated industries
summary
Demonstrates a reproducible secure ML stack built on SageMaker Studio plus VPC endpoints, IAM condition keys, KMS keys, Service Catalog and AWS Organizations SCPs; live demos show a Studio domain with no internet egress, a VPC endpoint policy blocking S3 uploads, an IAM deny that prevents training unless subnets/security groups are supplied, and an SCP that denies training unless model artifacts use a customer KMS key. Claims 54% TCO savings over three years and the demos run successfully.
AI-written summary · about
chapters (20)
- 0:00Start
- 1:14automate the creation and provisioning of these secure environments
- 4:02isolating your environment using a virtual private cloud
- 5:07need to set up vpc endpoints inside of your vpc
- 6:47attach fine-grained policies to specific user profiles
- 7:34modify the vpc endpoint
- 8:49adding s3 bucket policies
- 10:31pass in the security groups and subnets directly to your training
- 11:12create the sagemaker hosted endpoint
- 13:40create user identities
- 17:07preventing data exfiltration
- 18:22specify your own kms keys to encrypt
- 19:02prevent data exfiltration
- 21:00automate the provisioning of these secure environments
- 22:26trace the lineage of your training processing or your batch jobs
- 22:55using tagging as a fine grain mechanism for access control
- 25:25discuss service control policies
- 26:19jump into the studio domain in a child account
- 27:24encrypt the model artifacts
- 29:06build and scale your secure machine learning environments
related talks
46mBuild ML environment for regulatory customers
26mRunning Secure AI/ML Experiment in a Unified MLOPS-DevSecOps Pipeline
34mLat Designing and Building Complex Machine Learning....
55mBuilding a secure MLOps pipeline, featuring PathAI
20mIndustrial-Scale Machine Learning
44mPragmatic Machine Learning in the Cloud
25mDesigning an AWS Well Architected Framework for AI/ML
21mEnd-to-end secure ML Development