Securely build generative AI apps & control data with Amazon Bedrock
summary
Amazon Bedrock exposes Amazon Titan plus Anthropic, AI21 and Stability models through one API, enforces TLS 1.2 and AES-256/KMS encryption, logs KMS and API calls to CloudTrail, and isolates customer inputs so no data used to improve models. Single-tenant endpoints allow single-tenant fine-tuning via SageMaker using training data pulled from a customer S3 (ENI/VPC access shown), while multi-tenant endpoints are read-only. The console keeps an encrypted prompt history, CodeWhisperer speeds developer POCs, and private connectivity uses private VPC endpoints.
AI-written summary · about
chapters (21)
- 0:00Intro
- 1:21Today's agenda
- 3:30Generative Al is powered by foundation models
- 5:17How foundation models differ from other ML models
- 11:31Amazon Bedrock supports leading FMs
- 14:18Isolated customization to drive differentiation
- 15:09Data privacy and localization
- 16:06Data security
- 19:49Configurable security controls
- 24:05Client connectivity
- 27:25Multi-tenancy inference
- 29:47Single-tenancy inference
- 31:36Model fine tuning
- 33:37Complete architecture overview
- 34:27AWS Identity and Access Management
- 35:29IAM/SCP-Example deny policy
- 36:14Security in the model: challenges and opportunities
- 44:13Example: Amazon CodeWhisperer
- 48:35More models on Amazon SageMaker JumpStart
- 50:11How to use JumpStart for building with FMs
- 51:39Start your generative Al journey today
related talks
21mProtecting data in generative AI applications with Amazon Bedrock
01h 00mSecurely accelerating generative AI innovation
20mDeep dive into Amazon Bedrock security architecture
24mBuilding GenAI applications with Amazon Bedrock
20mData is Your Differentiator: Building Secure and Tailored AI Systems
16mAccelerating privacy & security in AI with Amazon Bedrock and Tines
40mSecure Your GenAI Applications
20mBuilding a secure end-to-end generative AI application in the cloud