The Adversarial Robustness Toolbox (ART) thru 1.20.1 contains an insecure deserialization vulnerability (CWE-502) in its Kubeflow component's model loading functionality. When loading model weights from a file (e.g., model.pt) during robustness evaluation, the code uses torch.load() without the security-restrictive weights_only=True parameter. This allows the deserialization of arbitrary Python objects via the Pickle module. An attacker can exploit this by uploading a maliciously crafted model file to an object storage location referenced by the pipeline, or by controlling the model_id parameter to point to such a file. When the pipeline loads the model, the malicious payload is executed, leading to remote code execution.

References

• https://nvd.nist.gov/vuln/detail/CVE-2026-31229
• https://github.com/Trusted-AI/adversarial-robustness-toolbox
• https://www.notion.so/CVE-2026-31229-35d1e13931888172863dcc20beeb6b70