Amazon ECS now shows live deployment monitoring for Linear, Canary, and Blue/Green strategies directly in the AWS Management Console—no more tab-switching between tools. You'll see deployment timelines, traffic shifts, health checks, and circuit breaker status all in one place, with deep links to CloudTrail when things break.
source: [aws/whats-new]
Learn how to transform AWS Bedrock Guardrails telemetry into Open Cybersecurity Schema Framework (OCSF) format within CloudWatch, giving your security team unified visibility into AI prompt injection attempts and data redaction events. Treat guardrail interventions like any other security alert—they're just as valuable for investigation as failed logins or network intrusions.
source: [aws/security-blog]
xAI's Grok 4.6 is now live on Amazon Bedrock, bringing a beefy frontier model built for long-running agents, coding tasks, and knowledge work. With a massive 500K token context window and four reasoning effort levels, it's ready to handle serious ML workloads through Bedrock's Converse API and cross-Region inference.
source: [aws/machine-learning-blog]
Amazon EC2 X8i instances are now live in the South America (São Paulo) region, packing custom Intel Xeon 6 processors with up to 6TB memory and 3.3x better bandwidth than the previous X2i generation. These SAP-certified powerhouses crush memory-intensive workloads like SAP HANA, PostgreSQL databases, and AI inference—delivering 50% higher SAPS performance and 88% faster Memcached compared to X2i.
source: [aws/whats-new]
Amazon SageMaker AI dropped 13 inference launches in the first half of 2026, covering both fully managed endpoints and HyperPod Inference deployments. The updates tackle real pain points: inference recommendations, capacity-aware instance pools, tiered KV caching, and disaggregated prefill-decode processing to squeeze better performance out of your ML models.
source: [aws/machine-learning-blog]
AWS just dropped the new AgentCore Runtime in Amazon Bedrock AgentCore, and it's a solid upgrade for your serverless agents. You're getting elastic memory that actually reclaims unused RAM during sessions (so you pay less), plus consistent cold starts around 2 seconds instead of the old 5-30 second mess—now live in us-east-1, us-east-2, us-west-2, eu-west-1, and ap-northeast-1.
source: [aws/whats-new]
Amazon SNS just quadrupled its message payload limit from 256 KiB to 1 MiB, so you can now push bigger data chunks through your topics without splitting payloads. This lands on both SNS Standard and FIFO, works with SQS, Firehose, and Lambda subscriptions, and is available across all AWS regions today.
source: [aws/whats-new]