Amazon Bedrock Enhances AI Capabilities with Security and Operational Features

AWS open-sourced Dogwood, a policy language that extends Cedar to govern sequences of agent tool calls by allowing rules to consider past actions. This development enables more complex control over AI agent workflows, addressing limitations of Cedar which only evaluates single requests.

AWS published a guide on integrating OpenAI-compatible endpoints on Amazon SageMaker AI with Amazon Bedrock AgentCore runtime to build agentic workflows. This integration allows specialized agents to use different models, including custom or cost-optimized ones on SageMaker, alongside managed models on Bedrock, within a single framework. The approach provides flexibility for model choice, cost optimization, and data residency for AI applications.

AWS published guidance on configuring AgentCore Observability for AI agents deployed outside of AWS environments, including on-premises or other cloud providers. This allows users to centralize monitoring for AI agents built with frameworks like Strands Agents, LangGraph, and CrewAI, regardless of their deployment location.

Amazon Bedrock AgentCore Browser Tool, combined with Strands Agents, provides a fully managed browser service for AI agents to interact with legacy web applications. This tool addresses the challenge of automating older systems that lack modern APIs and require human-like interaction, reducing manual data entry and errors in critical business operations.

Amazon Bedrock AgentCore can be used to build multi-agent systems for M&A due diligence, automating data gathering, analysis, and compliance checks. This approach addresses challenges like slow cycles, fragmented data, duplicated effort, and governance requirements in traditional M&A processes.

OneAdvanced, a UK enterprise software provider, implemented over 50 AI agents using self-hosted Llama 4 Maverick and Llama Guard 4 models on AWS to meet UK data sovereignty requirements. This approach allowed them to maintain data residency within the UK, which is critical for their customers in regulated industries.

Solv Labs developed an AI agent-payments workflow using Amazon Bedrock AgentCore payments, integrating its ORACLE policy engine and ICME PreFlight for compliance verification. This system enables auditable and governed transactions for AI agents, addressing the need for proof and accountability in autonomous financial operations. The solution allows enterprises to manage agent spending in regulated environments with full audit trails and pre-authorization.

OpenAI's Daybreak Red and Daybreak Blue models, designed for cybersecurity defense, are now available to eligible customers on Amazon Bedrock. This integration allows cyber defenders to use advanced AI for vulnerability analysis and remediation within AWS's controlled infrastructure.

Anthropic released a reference deployment guide for its Claude apps gateway on AWS, providing a self-hosted governance layer for enterprise use of Claude Code and Claude Desktop. This deployment offers centralized controls for authentication, model access, cost attribution, and spend enforcement for AI administrators.

nOps, a cloud optimization solution, transitioned its FinOps analytics to Amazon Bedrock AgentCore, reducing the time to ship FinOps agents by 75%. This change improved product delivery, response quality, and reduced operational complexity for managing cloud commitment optimization across AWS, GCP, and Azure.

Cohere Health developed Cohere Policy Studio using Amazon Bedrock AgentCore to convert static clinical policies into structured, machine-readable data. This aims to automate prior authorization processes in healthcare, which are currently manual and vary widely.

The TReNDS Center at Georgia State University implemented an automated root-cause analysis system using Amazon Bedrock, AWS Lambda, and other AWS services. This system detects errors in real time, enriches them with log context and source code, and provides AI-powered analysis to reduce manual investigation time.

Microsoft has launched a dedicated AI Gateway tier for Azure API Management in public preview, designed to govern AI models and MCP tools rather than traditional APIs. This new tier centralizes the management of multiple AI providers and models, offering features like token limits, quotas, and cost governance at the gateway level. It matters because it provides a specialized solution for managing the growing complexity of AI model deployments across various providers, enabling better control and cost management for platform and application teams.

AWS has introduced Runtime Instances, a new compute option within Amazon Bedrock AgentCore Runtime, providing persistent, managed infrastructure for complex AI agent workloads. This allows AI agents to run continuously for multiple days, access GPUs, and coordinate within shared sessions, addressing infrastructure challenges for production AI deployments.

AWS introduced Dogwood, an open-source policy language and reference interpreter designed to govern sequences of AI agent tool calls, rather than individual actions. This development allows for more complex policy enforcement, considering prior events and overall workflow, which is crucial for the safe and controlled operation of AI agents.

Amazon Bedrock Guardrails intervention data can now be routed to Amazon Security Lake, allowing security teams to analyze AI-related incidents alongside other security telemetry. This integration enables unified querying of guardrail events with identity, network, and application security data for comprehensive incident investigation.

Amazon Bedrock AgentCore now includes temporal policies, which allow for stateful authorization rules that evaluate AI agent requests based on their session history. This update addresses security challenges in AI agents by preventing issues like incorrect tool sequencing or data fabrication, which stateless policies cannot catch.

Amazon Bedrock's AgentCore gateway now supports rate limiting for AI traffic, allowing per-user control over requests, concurrent connections, and token throughput. This feature helps maintain the availability of downstream services by managing traffic spikes to tools, inference models, and agents.

AWS Kiro has replaced its multiple agent harnesses with a single architecture built around the Agent Client Protocol (ACP), allowing Kiro's clients to communicate with its agent through a shared protocol. This change enables developers to choose their coding tools and AI agents independently, standardizing the client-agent interface rather than using proprietary implementations.

AWS released guidance on integrating Codex with Amazon Bedrock, OpenTelemetry, and Amazon CloudWatch to provide visibility into coding agent usage. This integration allows organizations to monitor adoption, consumption, and reliability of Codex activity through an AWS-native view, without adding a centralized proxy.

AWS published a guide on how to configure Claude Code on Amazon Bedrock to enforce single-Region data residency, specifically for compliance requirements. This allows organizations to ensure that all prompts, completions, and intermediate processing remain within a designated AWS Region, addressing strict data residency mandates.

This article details how to use Agent Skills to manage the lifecycle of Amazon Bedrock Automated Reasoning policies from a coding agent. Agent Skills provide specialized knowledge and workflows to coding agents, enabling them to build, test, deploy, and validate policies more effectively. This approach allows for repeatable, reviewable, and code-driven policy management.

PDI Technologies created PDI Brew, a tool that allows non-technical employees to deploy multi-tenant web applications on AWS using natural language descriptions. This system utilizes Amazon Bedrock and AWS Lambda to automate the provisioning of AWS resources and integrate AI capabilities into the deployed applications.

LendingTree has developed a multi-agent AI mortgage assistant on Amazon Bedrock to help borrowers navigate the home-buying process. This assistant educates users, understands their financial situation, and provides tailored mortgage options through natural conversation, addressing the complexity of mortgage decisions.

Mobileye implemented an AI Support Agent using Amazon Bedrock AgentCore to automate routine internal ticket status inquiries, which previously required manual steps across multiple systems. This solution reduced response times by 90% and achieved over 95% accuracy, freeing skilled engineers from repetitive tasks and improving efficiency.

Amazon Bedrock AgentCore harness is now generally available, providing scaffolding for production AI agents with persistent memory and tools. A new open-source n8n community node integrates this capability, allowing users to build agents in n8n's visual editor without writing infrastructure code.

Amazon has made Web Search generally available on Amazon Bedrock, allowing foundation models to access current web knowledge for grounding responses. This integration removes the need for developers to integrate and maintain third-party web search providers, simplifying the process of grounding models and reducing hallucinations.

Amazon Web Services published a tutorial on deploying an automated web insight extraction solution using Amazon Bedrock AgentCore Browser, Amazon Bedrock for AI analysis, Amazon OpenSearch Serverless, and AWS Lambda. This solution addresses the challenge of manually extracting insights from numerous websites, especially those with dynamic content, by providing a resilient, AI-powered pipeline.

Amazon Bedrock now offers automatic policy refinement for its Automated Reasoning feature, automating the diagnosis and fixing of policy failures. This update addresses a significant friction point in policy development by proposing formal-logic fixes for failing tests, improving the efficiency of creating and validating policies.

Amazon Bedrock AgentCore Observability and Amazon CloudWatch can be used to identify and diagnose performance bottlenecks and memory issues in AI agents running in production. This helps maintain user trust and control costs by addressing slow response times and unbounded memory growth before they impact users.

Amazon Bedrock has launched Advanced Prompt Optimization, a new tool designed to automate the process of migrating and optimizing prompts across up to five different generative AI models. This feature addresses the manual effort and time previously required for prompt engineering, which often led to model lock-in and underperformance.

Google Cloud announced enhancements to its borderless Lakehouse, built on Apache Iceberg, enabling AI agents to access and act on data across AWS Glue, Databricks Unity, and Snowflake Horizon. This update allows for zero-copy, cross-cloud analytics and bidirectional interoperability, which matters because it reduces data movement costs and complexity for AI-driven workflows.

Microsoft introduced a reference architecture for routing large language model (LLM) agent traffic on Azure Kubernetes Service (AKS). This architecture optimizes cost and performance for agentic workloads by intelligently directing LLM calls to appropriate models and GPU resources.

Amazon published best practices for configuring Bedrock Guardrails in code generation workflows to prevent issues like throttling, increased costs, and latency. These guidelines address the unique throughput characteristics of AI-powered coding assistants, which generate long streaming outputs and handle concurrent developer sessions. Proper configuration helps detect and block unsafe code patterns and sensitive information while maintaining efficiency.

Motorway and AWS Prototyping and AI Customer Engineering (PACE) developed an evaluation pipeline for AI agents, reducing incorrect results from 1 in 8 queries to 1 in 50. This pipeline combines the Strands Agents SDK with Amazon Bedrock AgentCore to improve reliability and detect issues faster in production AI agents.

Amazon Bedrock AgentCore optimization now provides insights to detect and prioritize behavioral failures in deployed AI agents, including those that do not generate error signals. This feature helps identify patterns affecting agent performance at scale, moving beyond individual trace inspection.

Amazon Bedrock Managed Knowledge Base now includes agentic retrieval capabilities, allowing it to handle multi-part, comparative, and exploratory questions more effectively. This enhancement addresses limitations of single-shot retrieval, which often fails to provide comprehensive answers for complex queries spanning diverse document types.

Amazon, Microsoft, and Google have introduced enterprise agent platforms that share a common architecture across their offerings. This shift indicates a move towards unifying the agent ecosystem, enhancing scalability and deployment for enterprises, similar to past advancements in platform as a service.

Amazon, Microsoft, and Google have developed enterprise agent platforms with a shared architectural framework. Their platforms—Amazon Bedrock AgentCore, Microsoft Foundry, and Gemini Enterprise Agent—feature core components like runtime and observability, signaling a shift from fragmented libraries to unified production agents.

Couchbase has implemented Amazon Bedrock to support its AI-powered Capella iQ, enabling flexibility in using multiple foundation models. This architecture enhances operational resilience and supports high availability across AWS Regions, accommodating increased enterprise usage.

AWS has launched Loom, an open-source platform for building and governing AI agents, featuring built-in security controls. It aims to provide organizations with a guideline for deploying AI agents at scale on AWS while addressing common operational challenges.

Amazon Bedrock has launched its Managed Knowledge Base, a streamlined solution for building enterprise search capabilities. This service simplifies the setup process for knowledge retrieval systems by integrating data ingestion, search infrastructure, and access control, eliminating the complexity of traditional setups.

The article details the development of a voice ordering system for restaurants using Amazon Bedrock AgentCore and Nova 2 Sonic. It addresses the issue of missed customer calls during peak hours, demonstrating how to create a telephony AI that can handle orders and bookings autonomously, thus improving operational efficiency.

Amazon has integrated computer vision, Strands Agents, and Model Context Protocol (MCP) for AI applications. This streamlined approach facilitates the processing of visual information and decision-making, enhancing accessibility for developers.

Thrad.ai implemented a multi-agent system using Strands Agents and Amazon Bedrock for automating social intelligence. This system enhances prospect discovery and personalized outreach, significantly improving efficiency for lead generation and sales outreach.

Amazon Nova Act's QA Studio now includes features for batch regression testing and CI/CD integration. This improves software delivery by allowing concurrent execution of multiple test cases, thereby streamlining the quality assurance process.

Bluesight has developed Prism, an agentic AI solution using Amazon Bedrock to enhance healthcare compliance across its product suite. This AI integration aims to streamline the complex compliance processes, reducing manual labor for hospitals managing drug pricing program compliance.

Amazon Bedrock AgentCore Gateway now supports OAuth 2.0 Token Exchange for multi-tenant AI agents, allowing better identity handling. This implementation helps maintain an audit trail by securely propagating user identity across different downstream APIs without compromising security.

A guide outlines how to create a semantic layer on AWS that integrates Stardog with Amazon Bedrock AgentCore for improved enterprise analytics. This solution enables generative AI agents to autonomously query disparate data sources without the need for ETL, streamlining data access and analytics.

KTern.AI has created AI agents designed for SAP transformations using Amazon Bedrock AgentCore, enabling autonomous orchestration of complex workflows. This shift aims to improve efficiency in enterprise-scale SAP projects, offering automation without the need for custom infrastructure.

Amazon Bedrock AgentCore and Mistral AI Studio simplify the development of a production-ready ecommerce MCP server. This integration reduces custom API work and security risks, allowing faster deployment of AI-driven customer experiences within ecommerce platforms.

Amazon Bedrock AgentCore can now leverage AWS WAF for enhanced security via two architecture patterns, addressing authentication issues during health checks. This integration enhances protection against web threats while managing authenticated traffic among existing AWS services.

Jamf has launched AI Governance to manage AI applications like Claude and OpenAI Codex on Macs using Amazon Bedrock. This development allows organizations to configure and control AI applications centrally, enhancing security and compliance across devices.

Amazon Bedrock has enhanced its platform with features to enforce zero data retention policies, allowing organizations to manage data handling after inference requests. This change is significant as it provides users with means to ensure compliance and control over how prompts and outputs are retained, particularly when sharing data with third-party models.

The tutorial outlines the creation of a serverless image editing agent using Amazon Bedrock's AgentCore harness. This simplifies the implementation by automating orchestration and tool management, allowing users to edit images via natural language requests with minimal custom code.

Amazon Bedrock AgentCore enables developers to build an AI support companion for AWS management. This solution reduces investigation time by consolidating multiple support tasks into a single conversational interface, streamlining operations for AWS support teams.

Amazon Bedrock introduces MiniMax models, offering three open-weight foundation models tailored for software engineering and AI tasks. These models ensure that organizations can maintain security and compliance while leveraging advanced AI capabilities.

Amazon Bedrock in AWS GovCloud (US) now supports NVIDIA Nemotron and OpenAI open-weight GPT OSS models. This enables U.S. government agencies to utilize advanced AI capabilities while ensuring compliance and data security within the requisite isolation boundaries.

Amazon has launched the Model Profiler, an open source tool that simplifies model selection in Amazon Bedrock. It aggregates model metadata from various sources into a unified interface, enabling teams to efficiently compare models' capabilities, pricing, and availability.

AWS is introducing new advanced AI models, including Anthropic’s Claude Fable 5, on its Bedrock platform while emphasizing security. The updates aim to provide customers with faster access to cutting-edge models without compromising safety and privacy.

Amazon has launched the AG-UI protocol to facilitate dynamic user interactions with AI agents on Bedrock AgentCore. This open protocol enhances the integration between various agent frameworks and frontend libraries, allowing for more sophisticated interactions in AI applications.

Amazon Bedrock now supports managed entitlements, allowing centralized subscription management for AI models across multiple AWS accounts. This feature reduces operational overhead for organizations using third-party models by eliminating the need for individual AWS Marketplace permissions in workload accounts.

Amazon Bedrock AgentCore now supports resource-based policies, enabling SaaS providers to centralize access control for multi-tenant AI applications. This feature allows distinct security configurations for different tenants within a shared infrastructure, improving compliance and management.