At DevDay 2026, OpenAI and Amazon presented Amazon Bedrock Managed Agents, powered by OpenAI, as a limited-preview way to deploy OpenAI-powered agents in AWS rather than sending customer data to OpenAI’s cloud. The distinction is that this is a managed environment for running agents, not just access to an OpenAI model through Bedrock.
8
13
- Partnership and operation: The service extends their April 2026 partnership, which had already announced OpenAI models, Codex, and Managed Agents for AWS in limited preview. It combines OpenAI’s agent harness with AWS infrastructure; AWS says inference and agent operation take place in the customer’s AWS environment, with customer data remaining in AWS.
15
16
7
8
- Runtime and controls: AWS describes a managed runtime with inference, memory, and skills, plus built-in identity and security. Each agent has its own identity and its actions are logged, giving enterprises a way to review what an agent did.
7
10
- Enterprise value: Bundling those capabilities is intended to reduce the work of turning an agent prototype into a governed production service, while letting customers use AWS infrastructure and integrations they already rely on. That is a product aim, not proof that deploying or securing agents has become automatic.
7
10
- Wider DevDay context: Bedrock Managed Agents sat alongside OpenAI’s broader announcements about agent-building tools, models, Codex, and privacy features, including a preview of Private Intelligence.
6
12 Its emphasis on identity and auditability was especially timely: days earlier, OpenAI disclosed that agents had leaked 53 ChatGPT users’ images online. AWS residency and logging address important enterprise concerns, but they do not by themselves establish that an agent cannot misuse access or disclose data.
1
7
Early customer: Insufficient evidence in the available sources to identify one reliably, so I would not name a company as a confirmed early customer.