The vulnerability CVE-2026-18618 affects the ml-metadata component used in Red Hat OpenShift AI (RHOAI). The root cause is an outdated statically-linked gRPC stack (version 1.46.3 from 2022) that is vulnerable to known HTTP/2 denial of service issues. An attacker with network access inside the cluster to the MLMD pod can exploit this by sending specially crafted HTTP/2 requests, causing the MLMD pod to crash or exhaust resources. This results in denial of service, disrupting all pipeline runs within the namespace. Red Hat classifies the impact as moderate and the overall severity as high (CVSS 7.5). The affected component is planned for removal in future product versions. Red Hat recommends mitigating the risk by strictly enforcing network policies to restrict access to the MLMD pod's port 8080, allowing only necessary internal components such as KFP v2 driver pods and DSP components to connect. No official patch or fix has been released yet, and users should monitor the vendor advisory for updates.

An attacker with network access to the MLMD pod within the cluster can cause a denial of service by exploiting HTTP/2 DoS vulnerabilities in the bundled gRPC stack. This leads to crashing or resource exhaustion of the MLMD pod, disrupting all pipeline runs in the affected namespace. There is no impact on confidentiality or integrity, but availability is significantly affected.

Red Hat currently has no official fix or patch available for this vulnerability. To mitigate the risk, enforce strict network policies that limit access to the MLMD pod's port 8080. Only allow inbound connections from essential components such as KFP v2 driver pods and designated DSP components. This reduces the attack surface and limits potential exploitation by in-cluster attackers. Monitor the Red Hat advisory for any future updates or patches.