Skip to content Skip to footer

AWS AI and Cloud Updates: What Changed for Production Apps, Observability and Security

What Happened

AWS announced the following changes on November 20–21, 2026. The updates span production telemetry, application APIs, databases, security controls and partner tools; preview features are identified below.

AI applications and data

  • On November 21, Web Search in Amazon Bedrock and Bedrock AgentCore added Spanish, French, German, Italian, Japanese and non-US English support. An optional country field accepts an ISO 3166-1 alpha-2 code to favor local sources; queries are processed in the Region called. Existing customers receive language support without code changes. [2]
  • DynamoDB vector search added numeric ranges, value lists, attribute-existence checks and AND, OR and NOT to inline filters in the SearchVectors API. S3 Vectors raised filterable metadata from 2 KB to 40 KB per vector; existing indexes can opt in through UpdateIndexMode. [18][24]
  • Amazon Transcribe introduced the synchronous TranscribeAudio API for audio from 0.5 seconds to one hour, accepting up to 25 MB inline or 500 MB from S3. It costs $0.0070 per audio minute, with PII redaction priced separately. [19]
  • SageMaker JumpStart added a Studio leaderboard comparing more than 70 LLMs on benchmark quality, speed and instance cost. Newly created Neptune Serverless instances use Graviton4, with AWS reporting up to 30% better performance at the same per-NCU price; existing clusters can upgrade in a later release. [21][8]

Observability and infrastructure

  • On November 21, CloudWatch Omni added collection of Azure resource logs, platform metrics, VNet flow logs and Application Insights traces, allowing teams to investigate AWS and Azure telemetry together. Setup connects a Microsoft Entra ID tenant to an AWS account and Region. [1]
  • CloudWatch extended the PromQL query window for OpenTelemetry metrics from seven to 30 days. The CloudWatch Observability EKS add-on and Helm chart version 8.0.0 or later also support application OpenTelemetry telemetry, Prometheus target scraping and cluster-wide telemetry filters. Built-in dashboards cover tools including Karpenter, KEDA, vLLM and Ray. EKS OpenTelemetry ingestion costs $0.08 per GB. [9][17][20][13]
  • Lambda now sends Event Function version state-change events to EventBridge, enabling deployment workflows without polling GetFunction. Trusted Advisor also added recommendations comparing eligible high-volume Event Functions with Lambda Managed Instances. [4][25]
  • RDS for Oracle Enterprise Edition added a Primary Endpoint that follows the current primary after a replica switchover. Replica First Patching upgrades a same-Region replica before switchover to reduce minor-upgrade downtime; standard replica charges still apply. [3][6]
  • Cognito multi-Region replication expanded to 16 additional Regions. AWS Network Firewall’s automated Internet firewall entered public preview with centralized VPC internet-gateway inspection and no VPC routing changes required. [7][5]

Administration and partner operations

  • AWS Quick added group assignments for custom permissions and usage limits on Professional and Enterprise plans. Enterprise prompt-safety guardrails are in preview for select customers, with detect-only and enforcement modes. [14][16]
  • On November 21, AWS announced a US East (N. Virginia) preview of Billing and Cost Management Data Hub, offering AWS FinOps Agent 13 months of hourly, resource-level cost data in the CUR 2.0 schema. The preview is free and refreshes daily. [15]
  • AWS Partner Central added account mapping, Demand Center campaigns, lead grouping and outreach recommendations, and agents for Partner Central and Marketplace support cases. Access and joint-planning eligibility vary by partner program. [10][11][12][22]
  • Paid-plan builders using the new AWS experience can receive SMS alerts when a project reaches 90% of its monthly spend limit; AWS pauses the project at the limit. [23]

Why It Matters to Businesses

The most immediate operational gains are fewer custom integrations and less polling: teams can consolidate cross-cloud incident telemetry, collect EKS application signals through an existing add-on, and react to Lambda state changes through EventBridge. The Oracle endpoint and patching changes offer a clearer path to replica switchovers without changing application connection strings. [1][17][4][3][6]

For AI applications, richer vector filters can enforce retrieval constraints during search rather than relying solely on post-processing. Country-aware web search improves source relevance for localized experiences, but does not establish source accuracy. [18][24][2]

Kimbodo Engineering Perspective

These are useful platform primitives, not automatic production improvements. We would prioritize changes that remove a known failure mode or operating burden: replace Lambda polling only where event-driven orchestration is needed, adopt the Oracle Primary Endpoint before a planned switchover, and enable EKS telemetry filters before expanding collection. [4][3][20]

We would benchmark retrieval quality, Neptune workload performance and model choices against application-specific tests rather than treating vendor benchmarks or “up to” performance figures as expected outcomes. Preview firewall and guardrail features warrant staged validation before they become critical controls. [18][8][21][5][16]

How We Would Implement It

  • Inventory and stage: Identify affected AWS accounts, Regions, EKS add-on or Helm versions, Oracle replicas and AI search paths. Separate generally available changes from previews and confirm regional support. [17][3][5][16]
  • Instrument deliberately: Upgrade the EKS observability add-on to version 8.0.0 or later in a test cluster, send application telemetry to the node-local OTLP endpoint, and configure namespace and metric filters before broader rollout. Set retention, access and ingestion budgets. [17][20]
  • Test application behavior: Add DynamoDB SearchVectors filter tests covering mixed AND, OR and NOT conditions; migrate an S3 Vectors index through UpdateIndexMode only after checking metadata size and query behavior. Validate country-aware search with representative languages and markets. [18][24][2]
  • Automate recovery paths: Route Lambda state-change events through EventBridge to a deployment workflow with failure handling. For Oracle, move clients to the Primary Endpoint, rehearse replica validation and switchover, and verify reconnection behavior. [4][3][6]

Risks, Costs and Security

Cross-cloud telemetry requires careful scoping of the Entra ID connection, selected Azure subscriptions and collected signals; logs and traces may contain sensitive data. For AI retrieval, filterable metadata that includes permissions needs tests against unintended disclosure, not just relevance tests. [1][18][24]

Control ingestion and downstream spend: EKS OpenTelemetry telemetry is billed at $0.08 per GB, invoked EventBridge targets retain their normal charges, Oracle replicas retain standard pricing, and Transcribe PII redaction is an add-on. A free feature or preview does not make the surrounding workflow free. [17][20][4][6][19][15]

Use Quick guardrails in detect-only mode to measure false positives before enforcement, and review CloudWatch Logs access to guardrail events. Treat the automated Internet firewall as a public preview until its coverage, policy behavior and rollback procedure have been tested against real traffic. [16][5]

Where Kimbodo Comes In

Kimbodo builds and operates this in production for businesses — see our AI Application Development practice, or Estimate My AI Application.

Sources

  1. [1] Amazon CloudWatch Omni now supports Microsoft Azure platform logs, metrics and traces
  2. [2] Web Search on Amazon Bedrock expands language support and country-aware search
  3. [3] Amazon RDS for Oracle now supports Primary Endpoint
  4. [4] AWS Lambda now publishes Event Function state change events to Amazon EventBridge
  5. [5] AWS Network Firewall introduces automated Internet firewall in public preview
  6. [6] Amazon RDS for Oracle now supports Replica First Patching for reduced downtime
  7. [7] Amazon Cognito extends multi-Region replication to additional AWS Regions
  8. [8] Amazon Neptune Serverless now delivers up to 30% better performance and smarter scaling
  9. [9] Amazon CloudWatch extends the PromQL query window to 30 days
  10. [10] Account Mapping and Planning now available in AWS Partner Central
  11. [11] Demand Center now available in AWS Partner Central
  12. [12] Lead groups and channel recommendations now available in AWS Partner Central
  13. [13] Amazon CloudWatch EKS Container Insights now offers built-in monitoring for open source Kubernetes tools
  14. [14] AWS Quick now supports group assignments for custom permissions and limit profiles
  15. [15] AWS announces Data Hub in AWS Billing and Cost Management (preview)
  16. [16] AWS Quick introduces enterprise guardrails (preview)
  17. [17] Amazon CloudWatch Observability EKS add-on now supports custom OpenTelemetry and Prometheus telemetry
  18. [18] Amazon DynamoDB vector search now supports new inline filter operators
  19. [19] Amazon Transcribe announces a synchronous transcription API
  20. [20] Amazon CloudWatch Observability EKS add-on now supports telemetry filtering
  21. [21] Amazon SageMaker JumpStart now offers a model leaderboard to compare models on quality, speed, and cost
  22. [22] AWS Partner Central Agents now help partners create and manage Partner Central and Marketplace support cases
  23. [23] Spend limits in the new AWS experience now support SMS alerts
  24. [24] Amazon S3 Vectors raises the filterable metadata limit to 40 KB per vector
  25. [25] AWS Lambda now provides cost optimization recommendations for Event Functions

Leave a comment

0.0/5

Engineering Founded & Customer Focused

KIMBODO LLC. © 2026. All Rights Reserved.