AWS adds secrets management from Secrets Manager to Bedrock AgentCore

Amazon Web Services (AWS) announced a new capability for the Bedrock AgentCore platform — developers can now connect their own secrets stored in the AWS Secrets Manager service to the AgentCore Identity module, instead of creating separate credentials within the agent platform itself. This allows companies to extend existing secret management processes to AI agents without building parallel security infrastructure specifically for AgentCore.

How the new integration works

AWS Bedrock AgentCore is a service for deploying and managing autonomous AI agents in Amazon's cloud, and AgentCore Identity is its module responsible for agent authentication and granting them access to external systems through so-called "credential providers." Before this update, such providers required separate secret configuration within AgentCore. Now you can specify an already existing, pre-configured secret from AWS Secrets Manager — Amazon's service for centralized storage of passwords, API keys, and tokens — and AgentCore will reference it directly, without data duplication.

What this gives security teams

The main advantage for information security teams: full control over the secret lifecycle remains where it was before. The company retains management of encryption, key rotation, replication, tags, and resource access policies — just as it would manage any other secret in Secrets Manager. This eliminates a typical risk when implementing new AI services: the proliferation of fragmented credential storage systems, each with its own access and audit rules, which are difficult to control centrally.

Key details of the update:

• Service — AWS Bedrock AgentCore, AgentCore Identity module.
• Secret source — AWS Secrets Manager.
• Under company control remain: encryption, rotation, replication, tags, resource access policies.
• Supports using a secret from another AWS account in the same AWS region.
• Cross-regional secret sharing is not supported.
• Secrets connected via external Secrets Manager connectors are supported — that is, integration with third-party secret managers is also possible.

Why this matters for enterprise AI

For large organizations that have already built governance processes around secrets — with mandatory password rotation, access logging, and rights segregation by department — the ability to connect agents to existing infrastructure without migration is not a cosmetic but a systemic improvement. AI agents, unlike classical applications, increasingly gain access to sensitive systems — CRM, internal APIs, financial services — on behalf of a user or organization, so the question of how exactly an agent authenticates and who controls its credentials becomes one of the key issues when scaling agentic AI in an enterprise environment.

The limitation on cross-regional secret use, meanwhile, reflects AWS's typical approach to data isolation by geographic region — companies operating in multiple jurisdictions will have to configure secrets separately in each region where their agents are deployed, which may require additional planning for multi-regional AgentCore implementations.

Secret management becomes especially acute precisely for AI agents, since, unlike static services, an agent can independently initiate multiple calls to different external systems within a single session, which increases the number of points where an access key can be compromised. The ability to connect to an already existing, years-refined secret rotation and audit process instead of creating a new isolated storage specifically for agents reduces the so-called "attack surface" and simplifies compliance with internal security policies and external regulatory requirements to which large organizations in the financial, medical, and government sectors are subject. This update also reflects a broader AWS trend: gradually turning the disparate Bedrock platform services into a unified ecosystem integrated with already familiar AWS infrastructure tools that companies have been using for years regardless of whether they use AI agents or continue to work with classical cloud services.