CVE-2026-133225: Security vulnerability
A flaw was found in Keycloak. When processing token requests using the User-Managed Access (UMA) grant, Keycloak fails to verify whether a target client or resource server is enabled before issuing an authorization token. An authenticated remote user can exploit this flaw to obtain valid Requesting Party Tokens (RPTs) for disabled clients, bypassing administrative access controls and permissions restrictions.
Other sources
After an administrator disables a client that has Authorization
Services enabled (a resource server), Keycloak still issues new Requesting Party Tokens (RPTs) that target that disabled client. The UMA grant resolves the target resource server by checking only that the client and its authorization settings exist, and never checks whether the client is enabled. Keycloak then adds the disabled client to the RPT audience and includes the user's granted permissions. This bypasses the administrator's action of disabling the client, because ordinary token issuance excludes disabled clients from the audience of new tokens. The resulting RPT remains usable by any resource server or policy enforcement point that validates the signature, audience and embedded authorization data locally. The affected endpoint is the token endpoint (POST /realms/{realm}/protocol/openid-connect/token) using the UMA grant (urn:ietf:params:oauth:grant-type:uma-ticket).
— Red Hat
Affected Software
Frequently Asked Questions
Who can exploit this issue?
An authenticated remote user can exploit it. The affected target is a disabled client that has Authorization Services enabled and acts as a resource server.
What conditions are required for exploitation?
The attacker must be able to make a token request using the UMA grant and have granted permissions that can be included in the Requesting Party Token. The disabled target client and its authorization settings must still exist.
Does disabling the client prevent new tokens from being issued for it?
No. For UMA grant processing, Keycloak can still issue new Requesting Party Tokens with the disabled client in the token audience and the user's granted permissions.
When is an improperly issued token usable?
The resulting token remains usable by a resource server or policy enforcement point that validates the token signature, audience, and embedded authorization data locally.