AWS AI Updates: October 16, 2026

1. Meta’s Muse Spark Open Arrives on SageMaker JumpStart

AWS. Meta’s Muse Spark Open, a coding-optimized open-weight reasoning model, is now available in Amazon SageMaker JumpStart. The mixture-of-experts model has 816B total parameters with 42B active per token, a 1M-token context window, and accepts text, image, document, audio, and audio-video input with text output. JumpStart offers a BF16 variant that runs on ml.p6-b300.48xlarge and an NVFP4 quantized variant that needs roughly a quarter of the memory and runs on ml.p5e.48xlarge. Endpoints run in the customer’s own account and support VPC attachment, KMS-encrypted storage, and container network isolation. Source

2. Amazon MWAA Adds Four AI Agent Skills Through the Agent Toolkit for AWS

AWS. Amazon Managed Workflows for Apache Airflow (MWAA) now ships built-in agent skills through the Agent Toolkit for AWS, usable from any AI coding assistant that supports agent skills. The four skills cover authoring workflows, debugging workflows, testing workflows, and upgrading MWAA environments, so a pipeline described in plain language can be turned into a DAG with task dependencies and error handling. The skills work with both MWAA Provisioned and MWAA Serverless, are installed through the aws-data-analytics plugin, and are available in all Regions where MWAA runs. Source

3. AWS Resilience Hub Can Launch DevOps Agent Investigations From Findings and Tests

AWS. The next generation of AWS Resilience Hub now integrates with AWS DevOps Agent. When a resilience test runs, Resilience Hub launches a DevOps Agent investigation that observes the test and returns a structured analysis of symptoms, root cause, impact, and observed recovery actions such as ARC Region switch. Users can also start an on-demand DevOps Agent task directly from a failure mode assessment finding, with the finding’s context preloaded, to investigate and build a mitigation plan. The integration is configured per service in the Resilience Hub console and is available in 15 Regions. Source

4. AWS Outlines “Trust in Depth” Framework for Granting Agents More Autonomy

AWS. A post by AWS’s Shawn Bice describes “Trust in Depth,” a layered approach to agent governance organized around four questions: Scope (before an agent runs), Know (at deployment), Bound (while it runs), and Observe (after each run). It maps each layer to AWS services, including Kiro and Strands for build-time scoping, Policy in Bedrock AgentCore with automated reasoning checks on plain-language policies, AWS Agent Registry and AgentCore identity for tracking agents, Bedrock Guardrails and the open-source Strands Box sandbox for runtime limits, and Amazon CloudWatch Omni for trace capture and built-in evaluators. A worked example shows AgentCore Gateway blocking out-of-order steps in an employee offboarding workflow until required prerequisites, such as a legal hold check, are met. Source