AWS AI Updates: July 1, 2026

1. Claude Sonnet 5 Now Available on AWS

AWS. AWS made Claude Sonnet 5, the first Sonnet model in Anthropic’s latest generation, available through both Amazon Bedrock and the Claude Platform on AWS. The model targets coding across large multi-file codebases, agentic workflows with precise tool use and error recovery, and knowledge work such as generating spreadsheets and structuring unstructured data, all at Sonnet-tier pricing. Teams using Bedrock keep data within AWS and can attach Guardrails and Knowledge Bases with regional data residency, while the Claude Platform on AWS route offers the native Anthropic experience with unified AWS billing and authentication. Source

2. SageMaker AI Halves Inference Scale-Out Time With Container Image Caching

AWS. Amazon SageMaker AI now automatically pre-caches container images on new instances during scale-out events, delivering up to 2x faster end-to-end scaling for generative AI endpoints. Previously each new instance had to pull the full image from Amazon ECR, adding several minutes of cold-start latency for large images in the 10 GB-plus range common to generative AI workloads. The feature requires no configuration changes, works with accelerator instance types, single-model endpoints, and inference component-based endpoints, and is available in all commercial regions that support SageMaker Inference. Source

3. SageMaker AI Adds Serverless Customization for Gemma 4 Models

AWS. Amazon SageMaker AI now supports serverless model customization for Google DeepMind’s Gemma 4 E4B and 31B models, using supervised fine-tuning, direct preference optimization, and reinforcement fine-tuning. The serverless approach automates infrastructure provisioning and training coordination so teams pay only for resources consumed per job rather than managing clusters, and it extends existing support across Nova, Nemotron 3, Qwen, Llama, gpt-oss, and DeepSeek families. Customization is available in US East (N. Virginia), US West (Oregon), Asia Pacific (Tokyo), and EU (Ireland), launchable from the Models page in SageMaker Studio or via the SageMaker Python SDK. Source

4. Security Hub CSPM Launches AI Security Best Practices Standard

AWS. AWS Security Hub CSPM introduced an AI Security Best Practices standard with 31 automated controls that check AI resource configurations for network isolation, encryption at rest and in transit, VPC placement, KMS key usage, private container registry requirements, and authorization. The controls span Amazon Bedrock and AgentCore runtimes, gateways, memory stores, and custom browsers, plus SageMaker notebook instances, endpoints, models, monitoring jobs, and feature groups, generating findings when resources deviate from recommended settings. The standard (identifier standards/ai-security-best-practices/v/1.0.0) is available in all regions where Security Hub CSPM operates, including AWS GovCloud (US) and China Regions, with a 30-day free trial. Source

5. Amazon WorkSpaces for AI Agents Reaches General Availability

AWS. Amazon WorkSpaces for AI agents is now generally available, letting AI agents securely access and operate desktop applications through managed WorkSpaces environments without requiring legacy applications to be modernized. A new MCP tool forwarding capability lets agents interact through direct Model Context Protocol calls rather than computer use tools, which AWS says improves accuracy, reduces latency, and lowers cost. The service supports real-time session control for operator oversight, domain-joined fleets with Active Directory integration, and any agent framework that speaks MCP, with pricing based on active session time. Source