AWS AI Updates: June 3, 2026

1. Bedrock AgentCore Identity Supports Bring-Your-Own-Secrets via Secrets Manager

AWS. Amazon Bedrock AgentCore Identity now lets customers reference their own AWS Secrets Manager secret ARNs directly in Credential Providers, replacing the previous reliance on service-managed secrets. Teams can apply their own tagging, customer-managed KMS keys, rotation schedules, and resource policies to agent credentials, keeping governance entirely within existing security tooling. The capability is generally available across 14 AWS regions. Source

2. AWS Parallel Computing Service Adds a Production-Ready Deep Learning AMI

AWS. AWS Parallel Computing Service (PCS) now offers a PCS-ready Deep Learning AMI built on the Deep Learning Base GPU AMI (Ubuntu 24.04), with NVIDIA GPU drivers, the CUDA toolkit, EFA drivers, the Lustre client, EFS utilities, the PCS Agent, and multiple Slurm versions preinstalled and tested for compatibility. Supported Slurm versions activate automatically based on cluster configuration, giving teams a tested foundation for AI/ML training and HPC without manual driver and runtime setup. The AMI supports x86_64 and arm64, is available at no additional cost in all PCS regions, and receives regular security and driver updates. Source

3. Amazon Quick Adds VPC Connectivity for Private MCP Servers

AWS. Amazon Quick, AWS’s enterprise AI assistant, can now connect to privately hosted Model Context Protocol (MCP) servers over a VPC, reaching servers on EC2, Fargate, AgentCore, or other private compute without exposing them to the internet. When creating an MCP connector, users select a VPC connection and supply the server URL, letting teams query proprietary tools and data in natural language while keeping traffic network-isolated. The feature is available in all regions where Amazon Quick operates. Source