OpenAI and AWS Launch Bedrock Managed Agents to Keep Agent Data Inside Amazon’s Cloud

OpenAI and Amazon Web Services have launched Bedrock Managed Agents, a joint service allowing organizations to deploy autonomous OpenAI agents directly within Amazon's cloud infrastructure. Announced Tuesday at OpenAI’s DevDay 2026 conference in San Francisco, the integration is designed so that enterprise operational data remains entirely inside AWS throughout execution. The offering is currently in limited preview on the AWS product page, as reported by The Next Web (https://thenextweb.com/news/openai-bedrock-managed-agents-aws-devday).

The managed agent service builds on OpenAI’s agent framework and a customized version of its Agents API adapted to interface directly with native AWS resources. Under this architecture, AWS manages the runtime—handling model inference, memory, and skills—inside the customer's environment. All inference is processed via Amazon Bedrock, with Amazon’s AgentCore serving as the default compute environment. AWS noted that each agent maintains an isolated digital identity, records audit logs of every action, and will eventually incorporate native authorization policy enforcement, agent discovery, observability, and evaluation tools.

Salesforce has signed on as an enterprise customer. Joe Inzerillo, Salesforce’s president of enterprise and AI technology, stated that the offering enables organizations to pair OpenAI models and AWS infrastructure with Salesforce data and governance controls. “Data never leaves AWS, and security stops being the bottleneck between our customers and what comes next,” Inzerillo said on the AWS product page. The announcement follows Salesforce’s August move positioning Anthropic’s Claude at the center of its software lineup, as well as an AWS disclosure last week that nearly 90 percent of Amazon's early internal agent prototypes failed to ship.

Separately, OpenAI updated its underlying Agents API to support computer use, enabling developers to build agents that interact with software inside an OpenAI-hosted browser. The API also integrates multi-agent tools, tool search, and context compaction mechanisms used in Codex, with OpenAI managing the underlying infrastructure. Computer use is rolling out across the API, Codex, and ChatGPT Work for Pro 500 and Enterprise tier subscribers. OpenAI’s developer documentation specifies that users must approve each new website an agent navigates to, though individual in-browser actions do not require step-by-step confirmation, and sign-in credentials are routed via a segregated channel away from model context inputs.

The announcements come after OpenAI acknowledged on Monday that an agent malfunction resulted in private images belonging to 53 ChatGPT users being posted online. At DevDay, OpenAI also introduced always-on "dots" agents, revealed a partnership with Microsoft to deploy specialized dots to Agent 365, introduced a $500 Pro plan, and cut usage limits on its Pro 200 plan by half.