AWS this week added private certificate authority (CA) support to Amazon Bedrock AgentCore Identity. This means enterprise-deployed AI agents can connect to identity systems such as Keycloak and PingFederate running inside private networks, without switching to certificates trusted by default in public clouds. We see this as more than an isolated feature update: it is a critical piece in moving enterprise AI from “being able to call tools” to “being able to enter existing permission systems.”

What This Is

Many enterprises place identity management systems inside private networks and use private CAs to issue TLS certificates. AgentCore Identity can now read public CA certificates stored in AWS Secrets Manager or S3 and use them to verify an identity system’s discovery document, key endpoint, and token endpoint.

The CA certificate serves as a form of “proof of trust” for communications within the enterprise network and contains public key information, not the private key. In the article’s example, an agent runs through AgentCore Runtime, requests an OAuth 2.0 access token from an on-premises Keycloak instance through a credential provider, and then calls a permission-protected internal API. The entire process also relies on Amazon VPC Lattice to connect to the private network.

Industry View

The value of this update lies in reducing the work enterprises must do to rebuild identity authentication systems for AI agent integration. AWS has also demonstrated CA rotation and access control methods, showing that its goal goes beyond simply “connecting” to include certificate renewal and least-privilege access.

The risks, however, are equally clear. Enterprise identity systems typically connect to core business operations, employee permissions, and sensitive data. What an agent can do after obtaining a token depends on existing authorization rules, not merely the model’s capabilities. Private CA support lowers the barrier to integration, but it does not automatically solve excessive permissions, token leakage, or audit tracking. Supporting private CAs also adds complexity and cost to network architecture and certificate operations.

We therefore view it primarily as an infrastructure capability, not a sufficient condition for agents to safely enter enterprise networks. A truly mature solution must still answer three questions: how to minimize credential privileges, how to leave an audit trail for every call, and how to revoke access quickly when something goes wrong.

Impact on Regular People

For enterprise IT: Enterprises can more smoothly connect AI agents to existing internal knowledge bases and business systems. But they must simultaneously review identity permissions, network connectivity, and audit mechanisms; configuring certificates alone is not enough for a safe deployment.

For workplace professionals: Internal tools may gain agent entry points sooner. As employees use AI, they will increasingly come closer to a model of “calling systems on the employee’s behalf.” Enterprises may therefore strengthen operation logs, token expiration, and approval requirements for high-risk actions.

For the consumer market: The near-term impact is limited. Consumers will not directly handle CAs or credential configuration, but as enterprise agent services become more widespread, permission handling for applications such as customer service and shopping assistants will increasingly depend on this underlying security capability.