AWS AI Updates: October 7, 2026

1. SageMaker Studio Can Now Create and Manage HyperPod Spaces Without kubectl

AWS. Data scientists can now provision, start, stop, and open interactive development Spaces on SageMaker HyperPod EKS clusters from a guided UI in SageMaker Studio, choosing compute, namespace, storage, Task Governance quotas, and container images instead of using the CLI or kubectl. Spaces run JupyterLab or a VS Code-style Code Editor with persistent EBS storage, and SSH-over-SSM tunneling lets users attach a local VS Code. Admins install the Spaces add-on, attach three managed IAM policies, and set ExecutionRoleSessionNameMode: USER_IDENTITY so actions are attributed per user in EKS access entries and CloudTrail; startup takes minutes on cold clusters or 30 to 40 seconds with over-provisioning. Source

2. Reference Build Shows a Memory-Backed Personal Assistant on Bedrock AgentCore and OpenClaw

AWS. A new walkthrough builds “Sprout,” a gardening assistant on Amazon Bedrock AgentCore and OpenClaw, using AgentCore Memory for short-term events plus asynchronous extraction of user preference, semantic, and summary records that are filtered by metadata before being injected into the system prompt. It routes text to Claude Haiku 4.5 and vision tasks to Claude Sonnet 4.5, uses a Telegram webhook and EventBridge for scheduled reminders, and deploys serverlessly via CloudFormation for roughly $1 to $2 per month at light personal use versus about $35 per month for an always-on EC2 instance. AWS notes extracted long-term facts become retrievable in later sessions rather than immediately. Source

3. AWS Continuum, Formerly AWS Security Agent, Adds Continuous Pen Testing in CI/CD

AWS. AWS Security Agent has been renamed AWS Continuum for Penetration Testing and now offers, in public preview, penetration testing as a deploy-time step inside existing CI/CD pipelines, returning findings with severity, affected endpoints, and remediation guidance in pipeline output. Setup uses an auto-generated pipeline snippet that AWS says takes under five minutes, application context is bootstrapped automatically on first run without a prior full pen test, and fixes are re-tested automatically after remediation. Source

4. AWS Publishes a Multi-Tenant Governance Model for Shared SageMaker HyperPod Clusters

AWS. New guidance lays out governance for shared HyperPod clusters across four layers (organization, project, cluster, and workload) using SageMaker Unified Studio domain units, EKS access entries, RBAC, Slurm controls, and Task Governance priority classes and compute lending policies. It recommends centralizing clusters in a dedicated capacity account with scoped cross-account access, per-tenant namespaces and service accounts, default-deny network policies, and tenant-specific storage encryption. The post also separates authorization (who can submit work) from scheduling (when work gets compute) and treats utilization, wait time, and preemption metrics as a governance feedback loop. Source