CVE-2026-81207

IBM DataStage on Cloud Pak for Data 5.4.0.0 contains a server-side request forgery (CWE-918) in which any authenticated tenant — without requiring project membership or any role — can fully control the scheme, host, port, and path of an outbound HTTP fetch performed by the shared ds-canvas pod, and the fetched WSDL body is reflected verbatim in the response. An attacker triggers the flaw by issuing an authenticated request that causes the ds-canvas pod to fetch an attacker-chosen URL. Because the ds-canvas pod runs on the OpenShift overlay network, the attacker can reach co-tenant services, in-cluster Cloud Pak for Data APIs, and link-local (169.254.x.x) addresses, and the response reflection makes theft of sensitive data such as in-cluster API responses and secrets directly possible, with only minor integrity impact from GET-only side effects. Affected are organizations running DataStage on Cloud Pak for Data 5.4.0.0. No public proof-of-concept exists and the flaw is not in CISA KEV, so no exploitation is currently known.

What to do: Check IBM's security bulletin (CNA: [email protected]) for a fixed DataStage/Cloud Pak for Data release beyond 5.4.0.0 and upgrade when published. As interim mitigation, restrict network egress from the ds-canvas pod (OpenShift NetworkPolicy) so it cannot reach co-tenant workloads, in-cluster CP4D API endpoints, or link-local addresses, and consider limiting which authenticated tenants can trigger the WSDL fetch. Monitor ds-canvas egress logs for requests to unexpected hosts, unusual schemes/ports, or 169.254.x.x destinations, and audit in-cluster API tokens reachable from that pod.