Junglewise Threat Intelligence

CVE-2026-31230: Trusted-AI Adversarial Robustness Toolbox code injection in Kubeflow component

CVE-2026-31230 · Severity: critical · CVSS 9.8 · Published 2026-05-12

Technologies: Red Hat OpenShift AI (RHOAI), Trusted-AI Adversarial Robustness Toolbox. Vendors: Red Hat, Trusted-AI.

Executive brief

The Adversarial Robustness Toolbox (ART) is a Python library used by developers and data scientists to evaluate the security of machine learning models. A vulnerability in its Kubeflow component allows an attacker to execute malicious commands on the system running the evaluation. This could lead to a complete takeover of the machine, theft of sensitive AI models, or unauthorized access to corporate data.

Technical details

A command-line argument injection vulnerability exists in the Adversarial Robustness Toolbox (ART) through version 1.20.1 within the robustness_evaluation_fgsm_pytorch.py script of its Kubeflow component. The root cause is the unsafe use of the Python eval() function to parse string values provided via the --clip_values and --input_shape command-line arguments. An attacker who can control these arguments—for example, through a compromised pipeline configuration or automated script—can inject and execute arbitrary Python code. This can be exploited remotely without authentication, leading to full system compromise. Red Hat has also identified this as affecting OpenShift AI (RHOAI).

Affected products

  • Trusted-AI Adversarial Robustness Toolbox (ART) thru 1.20.1
  • Red Hat OpenShift AI (RHOAI)

Timeline

  • 2026-05-12: disclosed
  • 2026-05-12: advisory

References

Related threats