CVE-2026-31229 — Unsafe torch.load of Kubeflow pipeline model weights in ART

MITRE service request: 1988584

Status: RESERVED (pending a qualifying public reference per CNA Rules §5.3).

Official CVE description

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.

Summary

ART’s Kubeflow evaluation path downloads model.pt from attacker-influenced storage and calls torch.load with unsafe defaults. A poisoned checkpoint executes arbitrary pickle gadgets inside the evaluation worker.

Affected product and versions

• Product: Trusted-AI/adversarial-robustness-toolbox.
• Affected range (coordination record): through ART 1.20.1.

Technical details

• Surface: Pipeline parameters selecting model_id / COS object paths consumed by cos.fget_object before torch.load.
• Sink: torch.load(weights_filename, map_location=device) without weights_only=True.
• Impact: RCE on pipeline runner when malicious weights are loaded.
• Controls: Restrict write access to artifact buckets; validate model checksums; upgrade ART when patched.

Risk

High for shared Kubeflow clusters with writable artifact stores.

Remediation / workaround

• Pin torch.load(..., weights_only=True) where compatible; isolate evaluation pods.
• Vendor fix TBD.