This vulnerability in Red Hat OpenShift AI (RHOAI) affects the Data Science Pipelines component. It allows a restricted user or tenant to exploit improper authorization in the setDefaultServiceAccount function by specifying a more privileged ServiceAccount during a CreateRun request. The API server fails to perform a SubjectAccessReview for the requested ServiceAccount, enabling the attacker to bypass authorization checks. Consequently, the attacker can run workflow pods with elevated privileges, potentially accessing secrets and executing commands within other tenants' pods. The vulnerability has a CVSS 3.1 score of 7.1 (high severity) and is classified under CWE-639 (Authorization Bypass Through User-Controlled Key).

The vulnerability enables a restricted tenant to escalate privileges by running containers with more privileged ServiceAccounts than authorized. This can lead to unauthorized disclosure of sensitive information such as secrets and unauthorized command execution within other users' pods. The impact affects confidentiality (high), with limited integrity impact and no availability impact. There are no known exploits in the wild at this time.

Red Hat advises operators to configure an allow-list of ServiceAccounts that tenants can specify in their workflow run requests. Restricting the ServiceAccounts to a predefined, least-privileged set (e.g., the default 'pipeline-runner' only) prevents unauthorized privilege escalation. This configuration should be applied to the API server handling workflow run requests. There is no official patch or fix currently published; operators should apply this configuration mitigation until a fix is available. Patch status is not yet confirmed — check the vendor advisory for current remediation guidance.