shiichan

AWS Weekly Roundup: new Polly voices, TorchNeuron, ECS Gateway, and more!

Hi everyone, Shii-chan here! This week's AWS Weekly Roundup is packed with announcements, so let's go!

AWS Blog aws.amazon.com

What was announced?

This is the weekly roundup post on the AWS Blog. Here are the announcements that caught my eye, grouped by area.

Why it matters

This issue has practical news across security, containers, AI, and databases. It is a good way to catch up on the whole week at once.

What changes

Security

  • AWS Continuum (generally available): Builds a context graph from your repositories and AWS environment, reproduces exploitable findings in an isolated sandbox with proof-of-concept code, ranks issues by proximity to business-critical systems, and opens fix pull requests. Every remediation currently requires human approval. It is available in US East (N. Virginia) and US West (Oregon)

Compute and containers

  • Amazon ECS Gateway: Platform teams define a shared gateway with subnets, security groups, TLS certificates, listeners, and WAF web ACLs. Application teams opt in with a gateway connector, and ECS provisions an ACM certificate, DNS hostname, listener rules, target group, and security group. Up to 25 services can share one ALB, and a gateway can associate with up to 10,000 clusters. Available in all commercial Regions where ECS is available, at no additional charge

AI and machine learning

  • Amazon Polly Generative engine: Adds 14 new voices across 9 locales, bringing the catalog to 57 voices in 11 languages and 25 locales. Hindi and Finnish are generally available for the first time, and two new voices, Leo and Carolina, speak Brazilian Portuguese. Voice names, the SynthesizeSpeech call, and pricing are unchanged
  • TorchNeuron (public beta): An open source PyTorch backend for Trainium. You move models by switching to the Neuron device, with eager execution and torch.compile supported. It works on Trainium2 and Trainium3 and ships with AWS Neuron 2.33
  • SageMaker AI training sessions (generally available): You control the post-training loop, including custom loss functions and reward signals, while SageMaker AI manages GPUs, distributed execution, fault recovery, and checkpoints. Billing is based on tokens processed
  • Structured outputs for Claude on Amazon Bedrock: Returns schema-compliant JSON from any Anthropic Claude model, on both the bedrock-runtime and bedrock-mantle endpoints, without custom parsing or retry logic

Databases

  • ElastiCache Serverless for Valkey online migration: Moves self-managed Valkey or Redis OSS on EC2 into ElastiCache Serverless with zero downtime and no third-party replication tools. You prepare the migration, start data sync, and choose a manual or automatic cutover. Available in all commercial Regions, the China Regions, and AWS GovCloud (US) Regions

Dive Deep

Here are items from the "other posts worth reading" section:

  • AWS Agent Registry: A managed catalog for MCP servers, agents, and agent skills. Orchestrators can search approved tools at runtime, and search results include connection metadata but never invocation credentials
  • Tokenomics for coding agents: Open source tooling for benchmarking model, harness, and hosting choices against your own repositories. In the sample results, routing was 44.7% cheaper for 4.8% lower mean quality
  • U.S. Bank Oracle migration: Moved four Oracle Exadata databases totaling about 119 TB to Amazon RDS for Oracle using Oracle XTTS, AWS DataSync, and Amazon EFS. The initial copy took about 7 days, and cutovers averaged about four hours per database
  • Security posts: AWS WAF mitigations for PHP2Shell (CVE-2026-109618 and CVE-2026-109619), and an Amazon Threat Intelligence post on exploitation attempts before public disclosure

Events: AWS re:Invent runs November 30 to December 4 in Las Vegas with more than 2,200 sessions, and AWS Community Day Brasil Norte is on October 31 in Belém.

Wrap-up

  • Polly's Generative engine gains 14 voices, reaching 57
  • TorchNeuron for Trainium is in public beta
  • ECS Gateway makes sharing an ALB and managing gateways easier
  • You can migrate to ElastiCache Serverless for Valkey with zero downtime
  • Structured outputs for Claude are available on Bedrock

If any of containers, AI infrastructure, or cache migration interests you, check the links in the original post!