AWS and OpenAI launch Bedrock Managed Agents in preview, running OpenAI's Agents API inside AWS governance

Bedrock Managed Agents is the deepest OpenAI–AWS integration yet. Rather than simply hosting OpenAI models, AWS and OpenAI jointly adapted OpenAI's Agents API to run natively inside AWS. Agents optimized for OpenAI models execute entirely within the customer's AWS environment, using existing identities, permissions and governance controls.

The managed pieces are the parts teams usually build themselves: durable session memory, automatic state preservation between steps, and native Model Context Protocol server support for tool connectivity. Every tool invocation and state change runs under an IAM role and is recorded in AWS CloudTrail. For regulated enterprises that is the selling point: agent actions become auditable the same way any other AWS API call is. The preview covers major US Regions.

The launch fits AWS's week-long push to be the multi-vendor agent hub. Bedrock now hosts GPT-6 Sol and Luna, GPT-6.1 Sol, Claude Opus 5.5 and Sonnet 5.5, and Grok 4.7. AWS also shipped CloudWatch Omni for agent observability and the Strands harness. Bedrock AgentCore remains AWS's own model-agnostic agent runtime, so customers now face a choice between the two approaches.

Caveats: it is preview only, US-only, and optimized for OpenAI models, which may limit portability. Investors were unmoved; Amazon shares slipped about 0.28% as the market waits for revenue from the Bedrock expansion. Watch for GA timing, pricing, and whether Anthropic gets an equivalent managed-agents integration.