エージェント型 AI とレジリエンス、SageMaker 関連の更新が中心の日でした。AWS Step Functions は AgentCore を活用したエージェント推論ステップを追加し、SageMaker AI は AI エージェントのモデルカスタマイズ向けマルチターン強化学習を提供開始しました。SFT/DPO によるツール呼び出し精度向上や、Bedrock 上での自律運用構築、モデルクォータ申請と運用課題トリアージの自動化も紹介されました。Amazon Cognito のマルチリージョンレプリケーションによるアプリの回復力向上が News Blog で取り上げられ、ARC Region switch は Amazon Aurora のスケーリングと Amazon Neptune グローバルデータベースのフェイルオーバーに対応しました。Amazon Bedrock は OpenAI/Anthropic 互換 API 向けに再設計されたコンソールを公開し、GPT-5.4 は GovCloud (US-West) で一般提供となりました。その他 SageMaker Unified Studio の 12 言語対応やノートブックスケジューリング、ECS Managed Instances の Trainium/Inferentia 対応、Amazon MQ の欧州ソブリンクラウド（ドイツ）提供などが発表されました。日本語ブログは KMS/暗号化やオブザーバビリティのまとめ、ランサムウェア対策セミナー報告を扱いました。
エージェントAI: Step Functions の AgentCore 推論ステップ、SageMaker のマルチターン強化学習、SFT/DPO
レジリエンス: Cognito マルチリージョンレプリケーション、ARC Region switch の Aurora/Neptune 対応
生成AIモデル/コンソール: Bedrock の互換 API 向け新コンソール、GPT-5.4 の GovCloud 一般提供
SageMaker/コンピュート: Unified Studio の 12 言語対応、ECS Managed Instances の Trainium/Inferentia
地域展開: Amazon MQ が欧州ソブリンクラウド（ドイツ）で提供開始
日本語ブログ: KMS/暗号化、オブザーバビリティまとめ、ランサムウェア対策セミナー報告
Amazon SageMaker AI now offers multi-turn reinforcement learning (RL), a new serverless model customization technique for fine-tuning models on multi-step, agentic tasks. SageMaker AI model customization lets you adapt foundation models using techniques such as supervised fine-tuning, reinforcement learning from verifiable rewards (RLVR), and reinforcement learning from AI feedback (RLAIF), without the undifferentiated heavy lifting of building and operating your own training infrastructure. Multi-turn RL extends this by training models against your own agent environment and rewarding the full sequence of decisions an agent makes across a task, helping you specialize smaller, lower-cost models to match or exceed the task accuracy of larger general-purpose models on your target workload.
Training models that power agents to reliably complete multi-step tasks is complex and time-consuming, often requiring custom infrastructure that takes weeks to build. SageMaker's Multi-turn RL offering handles this for you. You can connect your agent running on Amazon Bedrock AgentCore Runtime for fully managed hosting, or on Amazon EKS, Amazon EC2, AWS Fargate, or any infrastructure using the framework of your choice. SageMaker AI manages the full training loop, from rollout orchestration and trajectory collection to training and checkpoint management. Built-in MLflow tracking lets you inspect agent trajectories, rewards, and traces. Evaluation jobs report reward, pass@k, and trajectory metrics so you can benchmark a model before deploying it to a SageMaker AI endpoint or Amazon Bedrock. Multi-turn RL runs as a fully serverless capability, so you pay only for the tokens processed, with no infrastructure to provision or manage.
Multi-turn RL is available today through SageMaker Studio and the SageMaker Python SDK as part of Amazon SageMaker AI model customization. Supported models include Qwen 3.6 27B, Nova Lite 2.0, GPT-OSS-20B and Gemma 31B in us-west-2, and Nova Lite 2.0, GPT-OSS-20B in us-east-1. To get started with multi-turn reinforcement learning in SageMaker AI, visit the Amazon SageMaker AI documentation.
AWS Config now supports 9 additional AWS resource types across key services including Amazon Bedrock, Amazon Bedrock AgentCore, and Amazon SageMaker. This expansion provides greater coverage over your AWS environment, enabling you to more effectively discover, assess, audit, and remediate an even broader range of resources.
With this launch, if you have enabled recording for all resource types, then AWS Config will automatically track these new additions. The newly supported resource types are also available in Config rules and Config aggregators.
You can now use AWS Config to monitor the following newly supported resource types in all AWS Regions where the resources are available:
Resource Types:
Amazon Elastic Container Service (Amazon ECS) Managed Instances now supports AWS Trainium and AWS Inferentia, purpose-built AI accelerators designed to deliver scalable performance and cost efficiency for training and inference across a broad range of generative AI workloads. Amazon ECS Managed Instances is a fully managed compute option designed to eliminate infrastructure management overhead while giving you access to the full capabilities of Amazon EC2. By offloading infrastructure operations to AWS, ECS Managed Instances helps you quickly launch and scale your workloads, while enhancing performance and reducing your total cost of ownership.
With ECS Managed Instances, you get the application performance you want and the simplicity you need. Now you can create an ECS Managed Instances capacity provider and select the desired accelerated instance types, including Inferentia2, Trainium1, and Trainium2, then add NEURON_CORE=all configuration to the ResourceRequirement section of your task definition. This will instruct Amazon ECS to launch the instances you’ve specified and place a single task per instance, automatically allocating all the resources of the accelerator to your workload for optimal performance.
To get started with ECS Managed Instances, use the AWS Console, Amazon ECS MCP Server, or your favorite infrastructure-as-code tooling to enable it in a new or existing Amazon ECS cluster. You will be charged for the management of compute provisioned, in addition to your regular Amazon EC2 costs. To learn more about ECS Managed Instances, visit the feature page, documentation, and AWS News launch blog.
Amazon SageMaker Unified Studio enhanced its global accessibility by introducing support for twelve languages across the user interface. Supported languages include English (American), Chinese (Simplified and Traditional), French, German, Indonesian, Italian, Japanese, Korean, Portuguese (Brazilian), Spanish, and Turkish. With this launch, data engineers, analysts, and data scientists across global teams can navigate, build, and collaborate in the language they are most comfortable with, reducing friction and improving productivity.
Your preferred language is automatically detected based on your browser’s default language settings. You can also set your preferred language by choosing ‘Language selector’ in your profile settings and selecting the language. The selected language applies across the entire SageMaker Unified Studio user interface.
This feature is available in all AWS Regions where Amazon SageMaker Unified Studio is available, in both AWS IAM Identity Center-based and IAM-based domains. To learn more, visit the Amazon SageMaker Unified Studio documentation.
Amazon Keyspaces (for Apache Cassandra) now returns an iterator position in the GetRecords response for change data capture (CDC) streams, indicating whether a consumer has reached the tip of the stream or whether additional records may be available. Amazon Keyspaces is a scalable, serverless, and managed Apache Cassandra-compatible database service that lets customers run Cassandra workloads on AWS without managing infrastructure. CDC streams capture row-level changes to Keyspaces tables so customers can integrate with downstream analytics, replication, and event-driven applications.
Previously, customers polled CDC streams at a fixed cadence regardless of whether new records were available, leading to inefficient resource usage and unnecessary CDC consumption costs. With iterator position, customers can now adapt polling frequency based on whether the iterator is at the tip of the stream or has records pending, lowering CDC consumption costs while maintaining timely data processing. The GetRecords response now includes an iteratorDescription structure with an iteratorPosition field that returns either AT_TIP or BEHIND_TIP, enabling customers to optimize their data integration pipelines and event-driven architectures.
This feature is available in all AWS Regions where Amazon Keyspaces CDC is supported. To use it, customers need to update to the latest AWS SDK. To learn more, visit the Amazon Keyspaces product page and see Working with change data capture (CDC) streams in the Amazon Keyspaces Developer Guide.
Amazon Application Recovery Controller (ARC) Region switch helps customers orchestrate the failover of their multi-Region applications to achieve a bounded recovery time in the event of a Regional impairment. Today, we are announcing three new execution blocks — the Amazon Aurora serverless scaling execution block, the Amazon Aurora provisioned scaling execution block, and the Amazon Neptune global database failover execution block — which automate database scaling and failover for multi-Region workloads.
Customers running Amazon Aurora global database in active-passive configurations typically maintain a scaled-down secondary cluster to minimize cost. During failover, they must manually right-size and scale the secondary cluster to handle production traffic before routing requests — adding critical minutes to recovery time. The new Amazon Aurora serverless and Amazon Aurora provisioned scaling execution blocks automate right-sizing and scaling the secondary cluster as part of the Region switch plan, so it's ready for production traffic when failover completes.
Customers running Amazon Neptune global database face a similar challenge: failover requires scripting or manually deciding whether to switchover or detach-and-promote depending on the outage type — all under the pressure of an active incident. The new Amazon Neptune global database failover execution block automates both planned switchover and unplanned failover scenarios within a single plan, eliminating custom scripting during recovery.
All three blocks support cross-account orchestration, enabling a single plan to coordinate database operations across multiple accounts and Regions. To learn more, read documentation of Amazon Aurora provisioned scaling, Amazon Aurora serverless scaling and Amazon Neptune global database failover
AWS Compute Optimizer now lets you extend the lookback period of your Amazon EBS volume and Amazon ECS rightsizing recommendations from the default 14 days to 32 days, at no additional cost. A longer lookback period allows Compute Optimizer to account for monthly utilization patterns, such as month-end processing, when generating rightsizing recommendations. This can help you make better optimization decisions for your workload, leading to better cost and performance outcomes.
AWS Compute Optimizer supports 32-day lookback periods for five types of recommendations: EC2 instance, EC2 Auto Scaling group, RDS database, EBS volume, and ECS service. You can set the lookback period at the organization, account, or resource level through the console, AWS SDK, or AWS CLI. This feature is available in all AWS Regions where AWS Compute Optimizer is available, except the AWS GovCloud (US) Regions and the China Regions. To learn more, see the AWS Compute Optimizer User Guide.
Amazon Bedrock now supports GPT‑5.4 from OpenAI in AWS GovCloud (US-West) — giving government and regulated industry customers access to OpenAI's most capable frontier model for professional work, backed by the enterprise-grade security and goverment compliance scope of AWS GovCloud (US).
GPT‑5.4 supports native computer-use capabilities, and deep reasoning across coding, documents, and multi-step agentic tasks — all running on Bedrock's high-performance inference engine with isolated queues and durable state for fault-tolerant workloads. Your data stays in-partition and is never used to train models.
For Regional availability of GPT-5.4 see the AWS Regions page. Read the launch blog to learn more, for documentation and a step-by-step walkthrough, see the Amazon Bedrock docs and the getting started blog.
AWS Step Functions now enables you to add AI agent reasoning steps to your workflow through an optimized integration with the managed harness (currently in preview) in Amazon Bedrock AgentCore. AWS Step Functions is a visual workflow service that orchestrates AWS services with built-in error handling, parallel execution, and human approval steps. The AgentCore harness lets you declare an agent through configuration where you specify the model, tools, and behavior. AgentCore provides the managed environment that runs the agent loop end-to-end.
With this integration, you can automate reasoning tasks in your workflow such as classifying a document or extracting elements from an unstructured form. You can run multiple agents in parallel or in sequence at different decision points in a single workflow and add human approval before critical actions. The workflow execution history shows agent input, output, token usage, and duration with links to agent turn details in Amazon CloudWatch, so you can trace and audit every agent decision. You can reuse an existing harness or create a new one directly from the Workflow Studio, the Step Functions visual builder. With per-invocation overrides such as the model, system prompt, and tools, you can adapt the agent to each workflow context without duplicating configurations. Agent context can be persisted across invocations using a session ID that works within or across workflow executions.
The harness integration is available in the following AWS Regions where the AgentCore harness preview is available: US East (N. Virginia), US West (Oregon), Europe (Frankfurt), and Asia Pacific (Sydney). Standard Step Functions pricing applies for workflow execution with no additional integration charges, and standard Amazon Bedrock and AgentCore pricing applies for model inference and associated AgentCore resources.
To learn more about adding agentic reasoning to your workflows, visit AWS Step Functions documentation.
Amazon SageMaker Unified Studio now enables you to schedule, parameterize, and orchestrate notebook runs directly from the notebook interface without managing external orchestration infrastructure. This makes it easier for customers to take notebooks from experimentation to production, automating recurring workloads such as daily reports, data quality checks, and model retraining.
You can trigger on-demand background runs on dedicated compute without interrupting interactive sessions and create scheduled or recurring runs. With notebook parameterization, you can reuse a single notebook across different inputs, for example, generating shipping performance reports for multiple carriers, by defining parameters and overriding their values per schedule or on-demand run. You can also orchestrate multi-notebook workflows using the Notebook Operator in the Workflows tool, chaining notebooks so that outputs from one run feed as inputs to the next. When a scheduled or background run fails, AI-assisted troubleshooting using SageMaker Data Agent helps you identify the root cause and suggests fixes directly in the notebook, reducing time to resolution. You can also use the Data Agent to create schedules and start notebook runs using natural language, without having to navigate. To get started, open a notebook in your SageMaker Unified Studio project, choose the menu on the Run all button, and select Run in background. To create a schedule, choose the schedule icon in the notebook header or ask the Data Agent to set one up for you.
You can use notebook scheduling in all AWS Regions where Amazon SageMaker Unified Studio is supported. To learn more, see the AWS blog and user guide.
Amazon SageMaker Data Agent, available in SageMaker Unified Studio now supports conversation history, enabling data practitioners to maintain continuity across analytical sessions. Data analysts and data scientists can now seamlessly reference previous agent-generated code, resume multi-step analyses, and review past troubleshooting interactions within their notebooks and Query Editor workflows.
With conversation history, you can pick up exactly where you left off by accessing a scrollable list of past conversations through the clock icon in the chat panel header. Each conversation includes auto-generated titles and timestamps for easy identification. Whether you're resuming complex multi-step analyses, reusing agent-generated code, or continuing troubleshooting from earlier notebook runs, conversation history keeps the context preserved. Data teams save time, eliminate rework, and move faster across concurrent projects, staying focused on insights rather than rebuilding context.
Conversation history is available in all AWS Regions where Amazon SageMaker Data Agent is currently available. To learn more about Amazon SageMaker Data Agent and how to leverage conversation history in your analytical workflows, visit the Amazon SageMaker product page or explore the Amazon SageMaker Unified Studio documentation.
AWS IoT Device Management adds MQTT session data to connectivity status API, enabling you to troubleshoot connectivity issues and audit connection patterns across your Internet of things (IoT) device fleet.
This launch brings AWS IoT Device Management's existing connectivity status API to full parity with AWS IoT Core's recently launched GetConnection API, enabling you to retrieve detailed connection and MQTT session information for the IoT device by its thing name. In addition to the connection status, timestamp, and disconnect reason already available, you now get visibility into MQTT session timeout and session expiry values, along with optional socket level details such as source and destination IP addresses, ports, and client VPC endpoint ID. Access to socket information is controlled through granular IAM policies, so you can restrict it to the teams that need it.
A key advantage of the connectivity status API over AWS IoT Core's GetConnection API is data retention. While GetConnection retains connection and session details for 30 minutes after a device disconnects, the connectivity status API stores this information indefinitely. This means you can investigate disconnect reasons, review session metadata, and troubleshoot issues long after a device goes offline.
This enhancement is available in all AWS regions where AWS IoT Device Management is supported. AWS IoT Device Management only supports devices registered in AWS IoT Core Thing Registry. To learn more, visit the AWS IoT Device Management documentation and reference guide.
Amazon Bedrock is a fully managed service that provides secure, enterprise-grade access to high-performing foundation models from leading AI companies, enabling you to build and scale generative AI applications. Today, Amazon Bedrock introduces a console experience designed for how customers actually build with foundation models: experiment, iterate, and scale. This is the same Amazon Bedrock service customers already use, with a refreshed workflow optimized for the bedrock-mantle endpoint, which supports the OpenAI Responses API, OpenAI Chat Completions API, and the Anthropic Messages API.
The new experience makes it simple to find the right model and move quickly from evaluation to production. Customers can browse the full Amazon Bedrock model catalog, including the latest Claude, GPT, and open-weight models, and compare them side by side on capabilities, modality support, context window, and applicable service quotas in a single view, removing the need to stitch together documentation, and limit calculators. Work is organized into projects, where customers can run evaluations and review usage insights in one streamlined workflow that mirrors the lifecycle of building a generative AI application. Each project also includes project-aware documentation: code samples, SDK snippets, and API references are automatically prefilled with the project's selected model ID, region, bedrock-mantle endpoint URL, and API key reference, and they update in place as customers change models or settings. Developers can copy a snippet straight from the console into their application and run it without modification.
To get started, sign in to the AWS Management Console, open Amazon Bedrock, and choose the new experience from the navigation. Create a project, pick a model, and begin sending requests through the bedrock-mantle endpoint using your existing OpenAI or Anthropic client libraries with an Amazon Bedrock API key. The new console experience is available in all AWS Regions where the bedrock-mantle endpoint is offered: US East (N. Virginia, Ohio), US West (Oregon), Asia Pacific (Jakarta, Mumbai, Sydney, Tokyo), Europe (Frankfurt, Ireland, London, Milan, Stockholm), and South America (São Paulo). To try the new experience, visit the Amazon Bedrock console.
You can now deploy Amazon MQ for RabbitMQ in the AWS European Sovereign Cloud (Germany) Region. This new independent cloud for Europe is located entirely within the EU, designed to help customers in regulated industries and public sector organizations meet their sovereignty requirements.
Amazon MQ is a managed message broker service that makes it easy to set up and operate message brokers in the cloud. Amazon MQ for RabbitMQ manages the provisioning, patching, and maintenance of RabbitMQ brokers, letting you focus on building applications without managing messaging infrastructure. You can migrate existing RabbitMQ workloads without rewriting application code and benefit from the same familiar APIs and protocols. Amazon MQ for RabbitMQ in the AWS European Sovereign Cloud supports RabbitMQ engine version 4.2 and Graviton3-based m7g instance types for high-performance messaging ranging from m7g.medium to m7g.16xlarge.
To get started, see the Amazon MQ product page or the Amazon MQ Developer Guide.
Amazon Cognito now offers multi-Region replication that automatically synchronizes user data, credentials, and pool configurations to a secondary AWS Region, enabling uninterrupted authentication during regional failovers without forced password resets—plus new support for customer managed KMS keys for encryption control.
AWS オブザーバビリティ関連リリースまとめの第1回へようこそ！2026年の最初の5か月間は、AWS オブザーバビリティにとって大きな変革をもたらした期間となり、 Amazon CloudWatch 、 AWS X-Ray 、 Amazon Managed Grafana 、 Amazon Managed Service for Prometheus にまたがって40を超えるリリースが行われました。この期間を特徴づける2つの大きなテーマは、統一された計装標準である OpenTelemetry 対応を強化したことと、オブザーバビリティを誰もが利用できるようにする AI 駆動のオペレーション です。EKS上でコンテナを実行している方も、複数リージョンにまたがるデータベースを管理している方も、AI 支援のワークフローを構築している方も、ここには役立つ情報があります。
大量のデータを暗号化する大規模アプリケーションでは、AES-GCM の暗号化限界の追跡や鍵のローテーションが課題になります。本記事では、AWS KMS と AWS Encryption SDK が派生鍵方式を用いて、暗号化のたびに一意の鍵を生成し、AES-GCM の呼び出し限界やデータ境界を自動的に処理する仕組みを解説します。鍵導出関数 (KDF) やノンスの活用により、手動管理を不要にする方法を詳しく説明します。
2026年5月21日、「ランサムウェアに備える『防御』と『復旧』— AWS で実現するセキュリティ対策」セミナーを開催しました。AWS サービスによる防御・検知・復旧の具体策と、パートナー3社によるエンドポイント・ネットワーク・運用支援をご紹介。各セッションの概要をレポートします。
日本の製造業のお客様に向けて、AWS の最新情報を毎月お届けするブログシリーズです。今号では、6月開催の AWS Summit Japan 2026 の製造業向け展示の見どころを中心に、Hannover Messe 2026 のブースレポート、トヨタ・Volkswagen 等のグローバル事例など5月の製造業×クラウド関連トピックを幅広くまとめています。
This post shows how to build a highly available Oracle database architecture using FSxN shared storage, Auto Scaling groups with dynamic AMI updates, and serverless orchestration to help reduce recovery times with current configurations.
In this post, we show you how to run a one-hour prioritization session with your stakeholders, plot competing initiatives on a shared matrix by cost and impact and turn the result into an actionable architecture backlog - using a framework called Tech Roadmap Prioritization (TRP).
In this post, you learn how to use Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO) together to improve the tool-calling accuracy of a small language model (SLM). The example uses Amazon SageMaker AI training jobs, so you can focus on training code instead of managing your own training infrastructure. You also learn how to evaluate tool-calling accuracy and compare a base model to several fine-tuned variants, so you can make data-driven decisions about model quality.
In this post, we look at how to use SOCI on publicly available Deep Learning AMIs and Containers, when to use the various SOCI modes provided by the tool, and how to quickly and efficiently use this tool in your workloads today.
In this post, we show you how to get started with NEXUS on Amazon SageMaker JumpStart, walk through the deployment process, and demonstrate how to run predictions against your enterprise datasets.
In this post, we introduce Amazon Bedrock Ops Alert, a three-layer automated monitoring solution that proactively detects operational issues, dynamically adjusts alarm thresholds, classifies alarms by category, automatically creates context-aware support cases, helps prevent duplicate cases when an unresolved case of the same alarm category is already active, and delivers contextualized notifications to AI SRE teams. We walk through the solution architecture and how you can deploy it in your own environment.
In this post, we introduce Amazon Bedrock Ops Alert, a three-layer automated monitoring solution that proactively detects operational issues, dynamically adjusts alarm thresholds, classifies alarms by category, automatically creates context-aware support cases, helps prevent duplicate cases when an unresolved case of the same alarm category is already active, and delivers contextualized notifications to AI SRE teams. We walk through the solution architecture and how you can deploy it in your own environment.