LoginSign Up
← Back to all sparks
A

AWS Machine Learning

AI-ASSISTANTS
Velocity10.0

Amazon Web Services' official AI/ML blog covering Bedrock, SageMaker, AgentCore, and Nova model updates.

aws.amazon.com/blogs/machine-learning ↗

AWS pours its blog into agentic Bedrock primitives and regulated-cloud model access

bedrockagentic-aimodel-availabilitygovcloudsagemakersecurity
◆Current state
The AWS Machine Learning feed is a firehose of blog posts, not a product changelog, so most entries are tutorials and customer showcases rather than shipped changes. Read for actual product signal, the recent cluster is clear: agentic infrastructure on Bedrock (AgentCore Memory, an A2A gateway pattern) and wider frontier open-weight model access.
◆Where it's heading
AWS is packaging Bedrock as the place to run and govern agents, not just call models: memory, agent-to-agent routing, and model selection tooling are all being fleshed out. The other throughline is regulated and enterprise deployment, with GovCloud model availability and fraud/phishing detection framed as first-class use cases.
◆Prediction
Expect more AgentCore building blocks and continued expansion of which frontier open-weight models are available in restricted regions. Note the caveat: velocity here reflects blog cadence, not release cadence, so treat the signal as directional rather than a shipping count.

◆Recent moves

  1. 21h ago

    How Amazon Bedrock catches AI-generated phishing

    A blog explainer on using Bedrock to flag AI-generated phishing; content marketing around the security use case rather than a shipped product change.

    View source ↗
  2. 21h ago

    Best practices for multi-turn reinforcement learning in Amazon SageMaker AI

    A best-practices post on multi-turn RL training in SageMaker. Useful guidance, but it documents technique rather than announcing a new capability.

    View source ↗
  3. 1d ago

    Run NVIDIA Nemotron and OpenAI GPT OSS models on Amazon Bedrock in AWS GovCloud (US)

    GPT OSS (120B/20B) and NVIDIA Nemotron models become available on Bedrock in AWS GovCloud, extending the agentic-and-regulated arc into data-residency deployments. A concrete availability change, not just a how-to.

    View source ↗
  4. 1d ago

    Building a serverless A2A gateway for agent discovery, routing, and access control

    A reference-architecture walkthrough for a serverless agent-to-agent gateway. It illustrates the agentic direction but ships as a tutorial, not a product feature.

    View source ↗
  5. 1d ago

    Structured memory filtering with metadata in AgentCore Memory

    Metadata filtering in AgentCore Memory adds real retrieval control for multi-agent and multi-tenant setups, fitting the push to make Bedrock a place to build stateful agents.

    View source ↗
  6. 1d ago

    HippoRAG: Neurobiologically inspired RAG using Amazon Bedrock, Amazon Neptune, and personalized PageRank

    A build-it-yourself RAG tutorial stitching Bedrock, Neptune, and Titan embeddings. Demonstrates the stack but is educational content, not a release.

    View source ↗