As a Cloud Security Architect and DevSecOps Lead, I'm pleased to present today's summary of critical AWS updates, focusing on new features, security implications, and essential Terraform Infrastructure as Code (IaC) availability.

Machine Learning Highlights (Amazon SageMaker)

Amazon SageMaker JumpStart now offers Qwen3.6-35B-A3B-NVFP4, an NVIDIA-quantized variant of Alibaba's Qwen3.6-35B-A3B. This Mixture-of-Experts model is optimized for agentic coding, multimodal reasoning, and long-context understanding (up to ~1M tokens via YaRN scaling), with only 3B parameters activated per token. Its NVFP4 quantization preserves fidelity for multi-step agent pipelines. Additionally, Wan2.1-T2V-1.3B-Diffusers is available for efficient text-to-video generation on consumer-grade hardware, leveraging a diffusion transformer paradigm and novel Video VAE. From a security perspective, deploying these advanced models requires robust IAM policies for SageMaker execution roles, ensuring least privilege. The extensive context window of Qwen3.6 necessitates careful prompt engineering and input validation to mitigate risks like prompt injection and data exfiltration. Encryption of data at rest (using KMS) and in transit is crucial for protecting sensitive model inputs and outputs. Architecturally, leveraging SageMaker's managed infrastructure offloads security patching and maintenance, but model validation for quantized versions is essential to ensure quality and prevent unintended biases or vulnerabilities. *Terraform Support*: Supported in Terraform AWS Provider: aws_sagemaker_endpoint_configuration in v5.82.0+.

SageMaker JumpStart expands its foundation model portfolio with Google DeepMind's Gemma-4-31B-it-assistant and NVIDIA's Gemma-4-31B-IT-NVFP4. These models bring the flagship Gemma 4 31B dense architecture, with the assistant-tuned variant supporting multimodal reasoning, coding, and agentic workflows across text and image inputs (including video frame sequences) with a 256K-token context window. The NVIDIA-quantized version (NVFP4) significantly reduces memory footprint (by 68%) and achieves ~2.5x faster inference while maintaining high quality. Security considerations include rigorous input validation for multimodal inputs to prevent malicious content or data exfiltration. The large context window demands secure data handling practices. Compliance requirements for data residency and privacy must be met, especially when processing diverse data types with these models. Leveraging SageMaker's built-in security features like VPC isolation and KMS encryption is paramount. *Terraform Support*: Supported in Terraform AWS Provider: aws_sagemaker_endpoint in v5.72.0+.

Mistral AI's Ministral-3-3B-Instruct-2512 and Ministral-3-8B-Instruct-2512 models are now available on Amazon SageMaker JumpStart. The 3.4B model is designed for ultra-lightweight edge deployment with multimodal understanding, fitting in 8GB VRAM (FP8), supporting 256K-token context, vision analysis, multilingual instruction following, and native function calling with structured JSON output. The 8.4B model offers frontier-class reasoning with similar capabilities, fitting in 12GB VRAM (FP8). The native function calling feature is a significant architectural and security consideration. It enables powerful agentic workflows but introduces a critical attack surface. Implementing strict access controls, input sanitization, and thorough security reviews of the functions invoked by the model are essential to prevent privilege escalation or unauthorized actions. Compliance with data handling regulations is also vital, especially with multilingual and multimodal inputs. *Terraform Support*: Supported in Terraform AWS Provider: aws_sagemaker_endpoint_configuration in v5.84.0+.

Amazon SageMaker JumpStart now hosts granite-speech-4.1-2b for multilingual Automatic Speech Recognition (ASR) and Bidirectional Speech Translation (AST) across six languages. kanana-2-30b-a3b-instruct excels in bilingual Korean-English instruction following and agentic AI, utilizing MLA and MoE architecture with 128K context support. OpenFold3 provides all-atom biomolecular complex structure prediction for proteins, DNA, RNA, and small-molecule ligands. Security insights for these models include ensuring the integrity and confidentiality of speech data processed by granite-speech, which may contain sensitive information. For kanana-2's agentic AI capabilities, the same rigorous security posture for function calling as mentioned for Mistral models applies. OpenFold3, dealing with highly sensitive biomolecular data, demands stringent data governance, encryption, and access controls to maintain data integrity and comply with scientific research and intellectual property regulations. All deployments should leverage SageMaker's secure environment and IAM best practices. *Terraform Support*: Supported in Terraform AWS Provider: aws_sagemaker_endpoint in v5.82.0+.

Messaging Highlights (AWS End User Messaging)

AWS End User Messaging has introduced support for dynamic WhatsApp flows, enabling rich, interactive experiences directly within WhatsApp. Users can complete tasks like appointment booking, lead generation, or surveys using Meta starter templates or custom JSON schemas with UI blocks. A key architectural and security consideration is that these dynamic flows call customer HTTPS endpoints in real-time for live availability and personalization. This introduces a critical attack vector that *must* be secured with robust authentication (e.g., mutual TLS, API keys, OAuth), authorization, comprehensive input validation, and rate limiting on the customer's endpoints. Data privacy and compliance (e.g., GDPR, CCPA) are paramount, requiring careful handling of user data collected through these flows and secure integration with backend systems. All flow configurations and user interactions should be encrypted at rest and in transit, ideally using customer-managed KMS keys. *Terraform Support*: While specific resources for defining dynamic WhatsApp flows may be pending dedicated provider releases, the underlying infrastructure (e.g., KMS keys, IAM roles, Pinpoint applications) is supported by the Terraform AWS Provider in v5.72.0+.

Announcements in this summary