CVE-2026-18611: Data Science Pipelines Information Disclosure

CVE-2026-18611 Overview

A flaw was found in the Data Science Pipelines Operator that allows unauthenticated attackers to derive sensitive credentials generated by the operator. The operator uses a cryptographically weak pseudo-random number generator (PRNG) to produce MariaDB root and user passwords, as well as MinIO access and secret keys. An attacker with network access to the MinIO Route or MariaDB Service can predict these credentials and gain unauthorized access to pipeline artifacts and metadata. The weakness is tracked as CWE-338 (Use of Cryptographically Weak PRNG) and affects Red Hat OpenShift AI deployments that ship the operator.

Critical Impact
An unauthenticated network attacker can derive MinIO and MariaDB credentials, leading to full disclosure of pipeline artifacts and metadata.

Affected Products

Red Hat OpenShift AI (Data Science Pipelines Operator)
Deployments exposing the MinIO Route
Deployments exposing the MariaDB Service

Discovery Timeline

2026-08-10 - CVE-2026-18611 published to NVD
2026-08-11 - Last updated in NVD database

Technical Details for CVE-2026-18611

Vulnerability Analysis

The Data Science Pipelines Operator provisions backend services including MariaDB for metadata storage and MinIO for artifact object storage. During deployment, the operator generates credentials for these services on the fly. According to the advisory, the credential generation routine relies on a non-cryptographic PRNG rather than a cryptographically secure random source. Because non-cryptographic PRNGs produce output determined by an internal state that can be reconstructed from limited observations, an attacker who understands the generation algorithm and its seeding behavior can predict future or concurrently generated credentials.

Exploitation requires only network reachability to either the MinIO Route or the MariaDB Service. Once credentials are derived, the attacker authenticates as a legitimate service consumer, reads pipeline artifacts stored in MinIO, and queries pipeline metadata from MariaDB. The impact is limited to confidentiality in the CVSS assessment, but the disclosed data can include model training inputs, output artifacts, and orchestration metadata.

Root Cause

The operator invokes a weak PRNG, such as a language-standard math/rand-style generator, instead of a cryptographically secure interface like crypto/rand. Seeding derived from predictable values, such as the current time, further reduces the search space required to reproduce generated secrets.

Attack Vector

An unauthenticated attacker sends network requests to an exposed MinIO Route or MariaDB Service endpoint. The attacker reconstructs likely credential candidates by replicating the PRNG algorithm and iterating over plausible seed values. Successful authentication returns pipeline artifacts and metadata to the attacker.

No verified public exploit code is available. See the Red Hat CVE Analysis for CVE-2026-18611 and Red Hat Bug Report #2510299 for vendor technical details.

Detection Methods for CVE-2026-18611

Indicators of Compromise

Successful authentication events to MinIO or MariaDB from source IP addresses outside expected pipeline execution ranges.
Bulk read operations against MinIO buckets that store pipeline artifacts, including unusual GetObject or ListObjects volume.
MariaDB connections from clients that do not match the Data Science Pipelines Operator service account patterns.
Externally reachable Routes or Services for MinIO and MariaDB where cluster policy expects internal-only exposure.

Detection Strategies

Audit OpenShift Route and Service objects created by the Data Science Pipelines Operator to identify externally exposed endpoints.
Correlate MinIO access logs against the identity of the pipeline orchestrator; flag authenticated sessions that do not originate from known pipeline pods.
Alert on repeated authentication failures against MariaDB or MinIO, which may indicate credential brute-forcing or seed enumeration attempts.

Monitoring Recommendations

Ingest MinIO server logs, MariaDB audit logs, and OpenShift API audit logs into a centralized SIEM or data lake for correlation.
Establish baselines for pipeline artifact access patterns and alert on deviations in volume, timing, or source workload identity.
Continuously inventory OpenShift Routes tied to data science workloads to detect unintended external exposure.

How to Mitigate CVE-2026-18611

Immediate Actions Required

Apply the fixes referenced in RHSA-2026:53261, RHSA-2026:53262, and RHSA-2026:53263 to update the Data Science Pipelines Operator.
Rotate all MariaDB root and user passwords and MinIO access and secret keys generated by affected operator versions.
Remove or restrict any OpenShift Routes that expose MinIO or MariaDB endpoints outside the cluster.

Patch Information

Red Hat has released updated Data Science Pipelines Operator packages through the security advisories listed above. The fixed operator generates credentials using a cryptographically secure random source. Consult the Red Hat CVE Analysis for CVE-2026-18611 for the specific product streams and versions covered by each errata.

Workarounds

Restrict network access to MinIO and MariaDB endpoints using NetworkPolicy resources that only allow traffic from pipeline workloads.
Delete externally facing Routes for MinIO and MariaDB where they are not required for the deployment topology.
Manually replace operator-generated credentials with high-entropy values sourced from a cryptographically secure generator until patches are applied.

bash

# Example NetworkPolicy restricting access to MinIO to pipeline pods only
# Adjust namespace and pod selectors to match your environment
apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata:
  name: restrict-minio-access
  namespace: data-science-pipelines
spec:
  podSelector:
    matchLabels:
      app: minio
  policyTypes:
    - Ingress
  ingress:
    - from:
        - podSelector:
            matchLabels:
              component: pipeline-runner