CVE-2026-12484 is a deserialization vulnerability in keras-team/keras version 3.15.0 where the method keras.layers.TorchModuleWrapper.from_config invokes torch.load with weights_only=False without requiring explicit unsafe opt-in. When called outside a SafeModeScope(True) context, safe deserialization is not enforced, allowing unsafe deserialization of attacker-controlled PyTorch pickle data. This can lead to arbitrary code execution if untrusted data is processed. The vulnerability is due to lack of enforced safe deserialization by default in this method.

An attacker who can supply untrusted Keras layer configurations to the vulnerable from_config method may achieve arbitrary code execution due to unsafe deserialization of PyTorch pickle data. This impacts confidentiality, integrity, and availability of affected systems. The CVSS 3.0 score is 7.8 (high), reflecting local attack vector, low complexity, no privileges required, user interaction required, unchanged scope, and high impact on confidentiality, integrity, and availability.

Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. Until an official fix is available, avoid processing untrusted Keras layer configurations with the from_config method or ensure that calls occur within a SafeModeScope(True) context to enforce safe deserialization. Do not disable safe mode or bypass safe deserialization checks.