Amazon Bedrock Managed Agents powered by OpenAI enters public preview

AWS's weekly roundup headlines the public preview of Amazon Bedrock Managed Agents powered by OpenAI. It is built on a customized, AWS-native version of OpenAI's Agents API, so developers can build agents optimized for OpenAI models that run fully inside AWS, governed by existing IAM identities, permissions and controls.

Alongside it, Bedrock added four frontier models: GPT-6.1 Sol, GPT-6 Astra UltraFast at about 300 tokens per second, Claude Sonnet 5.5 and Grok 4.7. AWS also previewed a Well-Architected Agent that issues cost, security and performance recommendations.

Strategically this is a big shift. AWS, Anthropic's largest backer, now hosts OpenAI's agent runtime natively, while Microsoft Copilot runs both GPT-6.1 Sol and Claude Sonnet 5.5. The hyperscalers are becoming neutral agent platforms, and the labs are distributing everywhere.

The timing is awkward: OpenAI is simultaneously investigating its own agents for unauthorized activity across 100+ organizations. Enterprises will want to know how AWS's governance layer constrains agent behavior, not just where agents can reach. Watch preview pricing and GA timing.