Bartek Chojnacki

AWS Cost Anomaly Detection now automatically monitors spending on third-party foundation models like Anthropic Claude running on Amazon Bedrock, with zero setup required. When your generative AI costs spike unexpectedly, you'll get instant alerts with a breakdown showing exactly which service, account, region, or usage type caused the jump—available across all AWS commercial regions except GovCloud and China.

Your AI agents pulling from DynamoDB, SaaS platforms, and internal knowledge bases need to know *who's asking*—otherwise they'll happily serve confidential data to anyone. Learn how to propagate user authorization context through Amazon Bedrock AgentCore so your agents respect access boundaries and don't become a data leak waiting to happen.

Amazon Bedrock now supports SpaceXAI Grok 4.6, a flagship model with a massive 500K context window and configurable reasoning levels, perfect for long-running agents tackling complex multi-step tasks like code analysis and research. You can access it across all AWS Regions with enterprise-grade security, monitoring, and cross-Region inference capabilities.

Amazon ion-java (Java library for Ion data format) has two critical memory-amplification DoS vulnerabilities: CVE-2026-75935 via declared-length preallocation and CVE-2026-75936 via compressed data expansion. Versions below 1.12.0 are affected and require immediate patching. If you're using ion-java in production, update now to protect against potential service disruptions.

OpenSearch Dashboards has a nasty input validation bug (CVE-2026-75897) that lets attackers hammer your system with oversized requests, causing denial of service. If you're running versions 1.3.0 through 3.7.0, you need to patch immediately—upgrade to 3.8.0 or apply the latest AWS service software update. AWS-managed OpenSearch users should grab the patched release available now.

AWS just dropped general availability for AgentCore payments in Amazon Bedrock AgentCore, letting your AI agents autonomously discover, access, and pay for APIs, MCPs, and content with minimal code. It's got enterprise-grade security, payment limits, and full observability—plus integrations with Coinbase and Stripe Privy wallets for microtransactions across multiple regions.

AWS just dropped batch instance termination for EC2 Auto Scaling, letting you nuke up to 100 instances with a single TerminateInstanceInAutoScalingGroup API call. Perfect for AI/ML training jobs and event-driven architectures that need to scale down fast—all instances validate atomically before termination, lifecycle hooks stay intact, and it's available everywhere at no extra cost.

Amazon ECR just bumped up the replication rules limit from 10 to 25 per registry, giving you way more flexibility for complex multi-region and multi-account setups. Now you can craft precise replication strategies across all AWS Regions—perfect for distributing container images globally with minimal latency or syncing to multiple prod and staging environments without juggling configs.

Amazon Bedrock just dropped cross-Region inference for OpenAI's GPT-5.6 models (Sol, Terra, Luna), letting you automatically route requests across AWS Regions for higher throughput and lower costs. You can now use Global inference for maximum capacity during traffic spikes, or stick with the new US Geo option to keep data processing within your region—all while using the same APIs (Responses, Converse, Chat Completions) and getting full observability through CloudWatch and Cost Explorer.