AWS Bedrock Agents Now Support Cross-Account Knowledge Base Access
New AgentCore capability allows enterprises to centralize RAG data while deploying AI agents across multiple AWS accounts.
AWS has introduced the ability to connect Amazon Bedrock agents to knowledge bases located in different AWS accounts. This update allows organizations to decouple data storage from application deployment, simplifying the architecture for retrieval-augmented generation (RAG) AI applications.
Under this new capability, Bedrock agents can securely access knowledge bases across account boundaries via the Amazon Bedrock AgentCore platform. The technical implementation relies on AWS cross-account IAM roles and resource-based policies—specifically the `bedrock:Retrieve` permission—to ensure that data access remains secure and governed while crossing account lines.
Breaking Down Data Silos
Previously, Bedrock knowledge bases were typically restricted to the same AWS account as the agent. For large enterprises with complex, multi-account structures, this limitation often created data silos, forcing teams to either duplicate datasets across accounts or consolidate all AI agents into a single, oversized account. This friction hindered the ability of large organizations to maintain a single source of truth for their corporate knowledge.
Scaling Enterprise AI
This shift is critical for enterprise scalability, as it enables a "hub-and-spoke" architecture. In this model, a central "hub" account manages the knowledge base and its associated vector store, while multiple "spoke" application accounts run the agents. This separation improves security and governance by isolating sensitive data management from the application layer, allowing data stewards to control access centrally without interfering with the deployment cycles of individual AI agents.
Governance and Implementation
As organizations move toward more mature AI operations, the ability to centralize knowledge while distributing execution is a key requirement for governance. While the core cross-account retrieval is now functional via AgentCore, developers must carefully manage the associated IAM and resource-based policies to prevent over-privileged access. By leveraging the `bedrock:Retrieve` permission, teams can ensure that agents only access the specific data subsets required for their function, maintaining a strict principle of least privilege across the organization's cloud estate. Future updates may further streamline how these cross-account permissions are orchestrated at scale to reduce the manual overhead of policy management.