Monday, November 16, 2026 Edition No. 240
Search
Daily News Fully automated · 7 min read

AWS AI Updates: November 10, 2026

Fully automated. Found, summarized and published by an automated AI pipeline. How posts are labeled

1. Claude Fable 5.2 Arrives on Amazon Bedrock and Claude Platform on AWS, With a 30-Day Retention Requirement

Amazon Bedrock made Anthropic’s Claude Fable 5.2 available on November 9 through Bedrock and Claude Platform on AWS. AWS describes it as Anthropic’s most capable generally available model for technical work, recommended for software engineering, scientific research and finance, and able to run unattended for hours to days and coordinate hundreds of subagents in Anthropic’s testing. Anthropic designates the model a Covered Model, so by default use requires retention of up to 30 days and human safety review, performed on Bedrock by Amazon personnel within the AWS boundary; customers set the aws_review data retention mode before invoking it, and eligible customers can use zero data retention until Enterprise Frontier Safeguards is available. If a cyber security query is declined, AWS says the request can be retried on Opus 4.8, and a declined biology query on Opus 5, but on Bedrock the retry is the caller’s code and a fallback credit token bills cached input as cache reads. The model ID in AWS’s example is global.anthropic.claude-fable-5-2, and AWS does not state pricing or context size in either post. Source Source

2. Fable 5.2 Comes to AWS GovCloud (US), While Mythos 5.2 Is Limited to Allow-Listed Organizations

Amazon Bedrock added Claude Fable 5.2 to AWS GovCloud (US) on November 9, aimed at regulated customers. The same date, AWS announced that Claude Mythos 5.2, which it describes as Anthropic’s most advanced model to date, is offered to eligible customers in GovCloud (US) for cybersecurity, software coding and complex reasoning. AWS calls it the first Anthropic model with powerful cybersecurity capabilities in GovCloud (US); it can find software vulnerabilities and demonstrate that they are exploitable, and access is limited to allow-listed organizations that go through Anthropic’s Cyber Verification Program. The announcements name neither the specific GovCloud Regions nor pricing. Source Source

3. Amazon Bedrock AgentCore Web Search Becomes a Standalone InvokeWebSearch API at $7 per 1,000 Queries

Amazon Bedrock AgentCore announced general availability of a Web Search API on November 10 that lets applications search the web with a single call to InvokeWebSearch, with no resources to create. Web Search was previously available only as an MCP tool through AgentCore Gateway. Results are ranked from an Amazon-operated web index and include a text extract plus, where available, the page title, URL and publication time, and requests can be filtered by domain and publication time. Calls are authorized with IAM, can run from a VPC over AWS PrivateLink, and are logged as CloudTrail data events. Pricing is $7 per 1,000 queries, the same as through Gateway with no gateway charges, in US East (N. Virginia), US East (Ohio), US West (Oregon), Asia Pacific (Sydney), Asia Pacific (Tokyo) and Europe (Ireland). Source

4. EKS Auto Mode and Karpenter Add Dynamic Resource Allocation for NVIDIA GPUs and EFA

Amazon EKS now supports Kubernetes Dynamic Resource Allocation (DRA) for NVIDIA GPUs and Elastic Fabric Adapter (EFA) network devices with EKS Auto Mode and Karpenter. Workloads declare ResourceClaim or ResourceClaimTemplate objects that select devices by attributes such as model or architecture, and Karpenter launches the lowest-cost EC2 instances that match. AWS lists GPU sharing across Pods, topology-aligned GPU and EFA allocation, and PCIe-aligned devices for distributed training and inference. Auto Mode includes the NVIDIA DRA driver and the EFA DRA driver (DRANET), and it prevents DRA and device-plugin workloads for the same device type from sharing nodes to avoid silent oversubscription during migrations. Karpenter support covers all Kubernetes versions, Auto Mode requires Kubernetes 1.36 or newer, and it is available in all Regions except GovCloud (US) and China. Source

5. AWS Nitro Sandboxes Offer Hardware-Isolated Pods, Tasks and VMs, With AI Agent Code Execution as a Named Use Case

AWS introduced Nitro Sandboxes on November 9, lightweight VMs managed by the Nitro Hypervisor that can host an Amazon EKS pod, an Amazon ECS task or a standalone EC2 VM, each in its own hardware-isolated environment. AWS positions them for running AI agents, safely executing LLM-generated code, and sharing one EC2 instance across several agentic workloads to raise utilization, as well as for untrusted, multi-tenant and regulated workloads. Cryptographic attestation lets secrets be released only to sandboxes that prove their integrity, and on EKS or ECS enabling them is described as a one-line change that keeps Pod Identity, VPC CNI, storage and CloudWatch working. On EC2, teams can bring their own AMI, orchestration and virtual machine monitor, with Amazon Linux and Firecracker natively supported. Nitro Sandboxes run on most Nitro-based EC2 instances in all commercial Regions at no additional charge. Source

6. NVIDIA Nemotron 3.5 Super VL, a 120B-Parameter Vision-Language Model, Is Deployable From SageMaker JumpStart

Amazon SageMaker JumpStart added NVIDIA Nemotron 3.5 Super VL on November 9. The model takes images, video and text and outputs text, pairing a hybrid Mixture-of-Experts language backbone of 120B total and 12B active parameters with a C-RADIO vision encoder, a 1-million-token maximum context and a built-in Multi-Token Prediction head for speculative decoding (not enabled in the JumpStart serving configuration). JumpStart offers an NVFP4 variant of about 75 GB on ml.g7.24xlarge (default) or ml.g7.48xlarge, and a BF16 reference variant of about 249 GB on ml.p5en.48xlarge intended for customization such as fine-tuning and distillation. NVIDIA targets document intelligence, GUI understanding for computer-use agents and video question answering, and the model is released under the OpenMDW License Agreement, version 1.1. Source

7. Bedrock Managed Knowledge Base Adds Single-Shot RAG and Retrieval-Only Modes to the Agentic Retrieval API

Amazon Bedrock Managed Knowledge Base now supports single-shot Retrieval Augmented Generation and retrieval-only requests through AgenticRetrieveStream, alongside the existing agentic retrieval that plans and runs multiple searches. The maxAgentIteration parameter now accepts 0 to 5, where one iteration equals one foundation model call: retrieval-only makes no model calls, a value of 1 retrieves once and generates one answer, and higher values allow up to n minus 1 planning rounds before the final answer. All three modes work with multiple knowledge bases, and applications can mix cheap lookups and multi-document questions through one API instead of separate code paths. AWS says single-shot RAG can reduce latency and model usage compared with agentic retrieval. Source

8. Amazon Quick Is Renamed AWS Quick, With Quick Sight Becoming AWS Quick Sight

AWS renamed Amazon Quick, which it calls an AI workspace built only for work, to AWS Quick on November 9, saying the name reflects the enterprise security and governance built into the service. Amazon Quick Sight, its business intelligence capability, is now AWS Quick Sight, and AWS states that existing customers need to take no action and that dashboards, datasets, analyses, settings and pricing are unchanged. The new name appears on the website, web, desktop and mobile apps, and in documentation. AWS lists HIPAA eligibility and certifications including FedRAMP, SOC 2 and ISO 27001, and states that customer data is not used to train other parties’ models. Source