CVE-2026-31228 is a remote code execution vulnerability in the Adversarial Robustness Toolbox (ART) Kubeflow component through version 1.20.1. The flaw is due to the use of the unsafe eval() function to dynamically evaluate user-supplied strings for the LossFn and Optimizer parameters without sanitization. This allows an attacker to provide specially crafted input strings containing arbitrary Python code, which is executed by eval(), resulting in full compromise of the system running the ART evaluation. The vulnerability is tracked under CWE-94 (Improper Control of Generation of Code). Red Hat has published an advisory confirming the vulnerability in their OpenShift AI deployment but currently has no available fix or mitigation that meets their product security criteria.

Successful exploitation allows an unauthenticated remote attacker to execute arbitrary Python code on the affected system, leading to complete compromise including full confidentiality, integrity, and availability impact. This can result in total control over the system running the ART evaluation, potentially allowing data theft, system manipulation, or denial of service.

As per the Red Hat advisory, no official fix or mitigation currently meets their criteria for deployment and stability. Users should monitor the vendor advisory for updates. Until a fix is available, avoid exposing the affected ART Kubeflow component to untrusted users or networks. Consider restricting access and applying compensating controls where possible. Engage with Red Hat support or Technical Account Managers for guidance.