CVE-2026-117574 is a critical vulnerability in ai-gateway-payload-processing used in multi-tenant Red Hat OpenShift AI environments. An authenticated attacker can specify another tenant's model identifier in a request, causing the service to inject that tenant's API keys without verifying authorization. This allows unauthorized use of another tenant's large language model resources and quotas. The flaw requires valid user credentials, limiting unauthenticated exploitation. No patch is currently confirmed available.