AWS Machine Learning Blog
AI news source. Articles are auto-selected and adapted by Hamidun News editors.
Latest publications
AWS presented two WAF protection patterns for Amazon Bedrock AgentCore Runtime
AWS published two ready-to-use architectural patterns for protecting AgentCore Runtime with AWS WAF: one with a Lambda proxy and one with a direct route to the ENI — both close off the backdoor through a resource policy.
Amazon Nova learned to forget: how rDPO reduces overcautiousness without quality loss
AWS introduced rDPO — a reverse preference optimization method that teaches Amazon Nova to precisely forget unwanted patterns without degrading quality.
Amazon Bedrock in AWS GovCloud supports NVIDIA Nemotron and OpenAI GPT OSS models
AWS added OpenAI GPT OSS open models (120B and 20B) and four NVIDIA Nemotron variants to Amazon Bedrock on the GovCloud (US) platform for U.S. government organizations.
Google DeepMind's Gemma 4 arrives on Amazon Bedrock: three models with MoE and multimodality
Amazon Bedrock has added three open-weight Gemma 4 models from Google DeepMind, featuring MoE architecture, built-in reasoning, and multimodal input support.
AWS Strands Evals automatically identifies root causes of AI agent failures and suggests fixes
AWS Strands Evals automatically diagnoses AI agent failures: identifies the root cause, builds causal chains, and indicates exactly what to fix—the system prompt or tool definition.
Amazon Bedrock Guardrails Gets a New API for Flexible Protection of Agentic AI Applications
AWS has launched InvokeGuardrailChecks, an API that lets developers embed safety checks at any point in multi-step agentic AI systems without pre-configuring guardrail resources.
AWS launched OAuth authorization for MCP servers in Bedrock AgentCore Gateway
Amazon Bedrock AgentCore Gateway added OAuth Authorization Code Flow — now every request to an AI assistant is tied to a real user identity.
AWS open-sourced Agent-EvalKit: systematic evaluation of AI agents in six phases
AWS released Agent-EvalKit, an open tool for structured testing of AI agents — it integrates with Claude Code, Kiro CLI, and Kilo Code and assesses six parameters.
Amazon Bedrock AgentCore Runtime: isolated microVMs for Claude Code, Codex, and Cursor
AWS has launched AgentCore Runtime — each AI agent gets its own isolated microVM with a persistent workspace and secure tool access via Gateway.
Amazon Bedrock adds cross-region routing for AI requests for EU customers
Amazon Bedrock now automatically distributes AI requests across AWS regions — European companies gain access to advanced models without breaching GDPR and the AI Act.
Amazon Quick generates incident RCA brief from a single prompt via New Relic and Asana
AWS shows how to build an AI agent on Amazon Quick that investigates an incident, generates an evidence-backed RCA brief, and creates an Asana task from a single prompt via New Relic MCP.
Amazon Bedrock AgentCore automates initial intake of insurance claims
AWS showed how an AI agent using Strands SDK analyzes an insurance case while Bedrock AgentCore Browser Tool independently enters data into insurers' portals — without operator involvement.
AWS and NVIDIA launched large-scale training for the Unitree H1 robot on SageMaker AI
Together with NVIDIA, AWS showed how to train control policies for the Unitree H1 humanoid in the cloud — through SageMaker HyperPod and Training Jobs with Isaac Lab.
AWS showed how to build an AI assistant for equipment repair on Bedrock AgentCore
AWS released a guide to building an AI assistant for diagnosing agricultural machinery on Amazon Bedrock AgentCore — with RAG, session memory, and the ability to describe the problem in your own words.
AWS launches Neuron Agentic Development to automate AI kernel development
AWS has unveiled Neuron Agentic Development, a suite of AI agents that automate compute kernel optimization for Trainium and Inferentia chips. Manual tuning is becoming a thing of the past.
Amazon Bedrock and HealthLake automate health insurance claims processing
AWS showed how to combine Bedrock Data Automation and AgentCore to automatically process health insurance claims and convert them into FHIR resources in HealthLake.
Amazon updates Bedrock AgentCore: agents gain access to enterprise knowledge and monitoring
AWS has expanded the Bedrock AgentCore platform: agents can now connect to internal and external knowledge bases, and teams have gained tools for production debugging.
Amazon Q launched autonomous agents — they run in the background while you do other things
Amazon Q has been updated: autonomous agents now run continuously in the background, the activity feed highlights what matters most, and unified search spans all enterprise data from a single query.
Amazon SageMaker AI removed mandatory S3 upload for asynchronous inference
Amazon removed the mandatory step of uploading input data to S3 for SageMaker asynchronous inference — data is now passed directly in the API request body.
Amazon SageMaker AI adds support for P-EAGLE to accelerate LLM inference in parallel
Amazon SageMaker AI now supports P-EAGLE, a parallel speculative decoding method that speeds up text generation in LLMs by 2–3x without degrading output quality.
AWS shows how data mesh becomes the foundation for agentic AI in production
AWS has published a technical guide to building a governed serverless data mesh — a secure foundation for agentic AI systems in enterprise production.
AWS open-sources Chaplin: self-service AWS Health analytics on Amazon Bedrock agents
Amazon has released open source Chaplin — an AI agent built on Amazon Bedrock and MCP for independently analyzing AWS Health events and infrastructure lifecycle changes.
Amazon SageMaker + NVIDIA Blackwell: How to Configure Model Training on P6-B200
AWS published a guide for setting up LLM training on SageMaker AI with NVIDIA Blackwell B200 GPUs, covering models from 1 to 64 billion parameters.
Stripe explains how AI agents help scale financial compliance
Stripe revealed the architecture of its production AI agent system for financial compliance — ReAct framework, task orchestration, human oversight, and prompt caching to reduce costs.
How Cara Creates Industry AI for Insurance Brokers in Partnership with AWS
Cara is a specialized AI assistant for corporate insurance brokers, created with AWS: RAG architecture, industry data, and measurable results in real enterprise brokerages.
Loka built a voice agent on Amazon Nova 2 Sonic with sub-second latency
Loka has disclosed the architecture of a voice agent built on Amazon Nova 2 Sonic — an AWS model that answers customers without a noticeable pause and sounds like a human operator.
Rocket Close Automated Property Rights Verification Using AWS Agentic AI
Rocket Close created the Supercharger agentic system based on AWS Strands Agents and Amazon Bedrock for automated property rights verification in real estate transactions.
AWS SageMaker adds full monitoring for LLM models in production
AWS expanded SageMaker AI with tools for full monitoring of LLM models. New Grafana dashboards track GPU utilization, processing speed, and response quality in real time.
AWS Launches AgentCore to Manage Dozens of AI Agents in Sales
AWS launched AgentCore — a platform to manage 20+ specialized AI agents. Instead of salespeople manually choosing between agents, the system automatically routes requests.
AWS Created NarrateAI — a Business Analytics Assistant on Amazon Bedrock AgentCore
AWS demonstrated how specialized AI agents and a two-layer architecture help scale business intelligence for SMGS divisions using Amazon Bedrock AgentCore.