AWS AI Updates: June 18, 2026

1. Amazon Bedrock Managed Knowledge Base Reaches General Availability

Amazon Bedrock. Bedrock Managed Knowledge Base is now generally available as a fully managed retrieval-augmented generation platform that removes the need to operate a vector database or retrieval infrastructure, with managed vector storage and automatic data syncing across six native connectors including Amazon S3, SharePoint, Confluence, Google Drive, OneDrive, and a web crawler. It adds advanced retrieval such as hybrid search, document ranking, and agentic retrieval with automatic query planning and response evaluation for multi-hop questions, plus support for multimodal sources spanning text, images, audio, and video. The service integrates natively with Bedrock AgentCore and ships with auto-generated permissions and built-in observability, available in US East (N. Virginia), US West (Oregon), Asia Pacific (Sydney, Tokyo), Europe (Dublin, Frankfurt, London), and AWS GovCloud (US-West). Source

2. Bedrock AgentCore Adds Production Optimization for Continuous Agent Improvement

Amazon Bedrock AgentCore. AgentCore now analyzes production traces across hundreds of sessions to surface failure, intent, and trajectory patterns, including silent behavioral failures, and ranks them by user impact through a new insights suite. A recommendations engine then generates data-grounded fixes to system prompts and tool descriptions, while batch evaluation against defined datasets and live A/B testing validate changes before full rollout, forming a closed loop of monitor, diagnose, fix, and verify. Insights are in preview across 13 regions and the batch evaluation, recommendation, and A/B testing capabilities are generally available in 14 regions, working across the AgentCore runtime, Lambda, EKS, and non-AWS infrastructure. Source

3. AgentCore Harness Becomes Generally Available for Configuration-Driven Agents

Amazon Bedrock AgentCore. The AgentCore harness is now generally available, letting developers define an agent through configuration that specifies the model, tools, skills, and instructions while the managed harness runs the orchestration loop and persists state across sessions. It provides isolated environments with filesystem and shell access, lets teams switch model providers mid-session without losing context, and integrates AWS security policies, organizational knowledge, and web search. Agents can later be exported to code-based implementations using Strands or the Claude Agent SDK when custom orchestration is needed, and the harness is available in all AWS commercial regions where AgentCore runs. Source

4. Bedrock AgentCore Enforces Guardrails at the Policy Layer

Amazon Bedrock AgentCore. AgentCore now integrates Bedrock Guardrails into its policy authorization framework, evaluating agent action outputs and gateway target inputs across tools, agents, and models in real time to block prompt injection, harmful content, and sensitive data exposure before they reach downstream systems. Because enforcement happens at the gateway perimeter outside agent code, controls apply consistently regardless of agent autonomy, and all evaluations are logged through AgentCore observability for auditing and optimization. Policies can be authored in natural language or as policy-as-code with consumption-based pricing, and the capability is available in US East (N. Virginia), Europe (London, Stockholm), and Asia Pacific (Sydney, Tokyo). Source

5. Web Search Arrives on Bedrock AgentCore for Grounded Agentic Retrieval

Amazon Bedrock AgentCore. AgentCore now offers Web Search, a managed tool that lets agents retrieve current web information while keeping data within AWS infrastructure with zero data egress. It combines Amazon’s web index with structured knowledge graph data for entity information and verified facts, returns ranked results with snippets, source URLs, titles, and publication dates, and is optimized to return high-value excerpts for strong intelligence per token. The tool is exposed as a built-in connector over the Model Context Protocol on the AgentCore gateway, removing the need to integrate external search providers and manage their authentication, and is available at launch in US East (N. Virginia). Source

6. Amazon S3 Vectors Cuts Query Charges by Up to 80% for Large Indexes

Amazon S3 Vectors. S3 Vectors now reduces data processing charges by up to 80% on queries against vector indexes containing more than 10 million vectors, lowering the cost of similarity search for large-scale AI, RAG, and semantic search workloads. The new pricing applies automatically with no application changes once an index crosses the 10 million vector threshold, making the savings structural rather than configuration-driven. AWS still recommends distributing vectors across multiple indexes for query performance, and the reduced pricing is effective in all AWS Regions where S3 Vectors is available. Source