Логотип exploitDog
Консоль
Логотип exploitDog

exploitDog

redhat логотип

CVE-2026-12484

Опубликовано: 19 июл. 2026
Источник: redhat
CVSS3: 7.8
EPSS Низкий

Описание

A vulnerability in keras-team/keras version 3.15.0 allows unsafe deserialization of attacker-controlled PyTorch pickle data through the public keras.layers.TorchModuleWrapper.from_config method. This method invokes torch.load(..., weights_only=False) without requiring an explicit unsafe opt-in, such as a safe_mode=False parameter. When called outside a SafeModeScope(True) context, the absence of an ambient safe mode state permits unsafe deserialization by default. This issue can lead to arbitrary code execution if untrusted Keras layer configurations are processed using this method. The vulnerability arises because the method does not enforce safe deserialization practices unless explicitly guarded by Keras safe mode.

A flaw was found in Keras, specifically within the keras.layers.TorchModuleWrapper.from_config method. This vulnerability allows an attacker to perform unsafe deserialization of PyTorch pickle data. By processing untrusted Keras layer configurations, a local attacker could exploit this flaw, leading to arbitrary code execution. This occurs because the method does not enforce safe deserialization practices by default, potentially allowing malicious code to run on the system.

Отчет

This Important vulnerability affects Red Hat OpenShift AI components utilizing Keras for model processing. A local attacker could achieve arbitrary code execution by providing untrusted Keras layer configurations, which are then unsafely deserialized by the keras.layers.TorchModuleWrapper.from_config method. This risk is present when users process external or unverified model configurations within their OpenShift AI deployments.

Меры по смягчению последствий

To mitigate this issue, ensure that Red Hat OpenShift AI environments only process Keras layer configurations from trusted sources. Avoid loading or processing any untrusted Keras layer configurations, as this vulnerability can lead to arbitrary code execution.

Затронутые пакеты

ПлатформаПакетСостояниеРекомендацияРелиз
Red Hat OpenShift AI (RHOAI)rhoai/odh-kserve-agent-rhel9Not affected
Red Hat OpenShift AI (RHOAI)rhoai/odh-kserve-controller-rhel9Not affected
Red Hat OpenShift AI (RHOAI)rhoai/odh-kserve-router-rhel9Not affected
Red Hat OpenShift AI (RHOAI)rhoai/odh-kserve-storage-initializer-rhel9Not affected
Red Hat OpenShift AI (RHOAI)rhoai/odh-modelmesh-runtime-adapter-rhel9Affected
Red Hat OpenShift AI (RHOAI)rhoai/odh-pipeline-runtime-tensorflow-cuda-py312-rhel9Affected
Red Hat OpenShift AI (RHOAI)rhoai/odh-pipeline-runtime-tensorflow-rocm-py312-rhel9Affected
Red Hat OpenShift AI (RHOAI)rhoai/odh-workbench-jupyter-tensorflow-cuda-py312-rhel9Affected
Red Hat OpenShift AI (RHOAI)rhoai/odh-workbench-jupyter-tensorflow-rocm-py312-rhel9Affected

Показывать по

Дополнительная информация

Статус:

Important
Дефект:
CWE-502
https://bugzilla.redhat.com/show_bug.cgi?id=2502670keras: torch: Keras: Arbitrary code execution via unsafe deserialization of PyTorch data

EPSS

Процентиль: 11%
0.00204
Низкий

7.8 High

CVSS3

Связанные уязвимости

CVSS3: 7.8
ubuntu
13 дней назад

A vulnerability in keras-team/keras version 3.15.0 allows unsafe deserialization of attacker-controlled PyTorch pickle data through the public `keras.layers.TorchModuleWrapper.from_config` method. This method invokes `torch.load(..., weights_only=False)` without requiring an explicit unsafe opt-in, such as a `safe_mode=False` parameter. When called outside a `SafeModeScope(True)` context, the absence of an ambient safe mode state permits unsafe deserialization by default. This issue can lead to arbitrary code execution if untrusted Keras layer configurations are processed using this method. The vulnerability arises because the method does not enforce safe deserialization practices unless explicitly guarded by Keras safe mode.

CVSS3: 7.8
nvd
13 дней назад

A vulnerability in keras-team/keras version 3.15.0 allows unsafe deserialization of attacker-controlled PyTorch pickle data through the public `keras.layers.TorchModuleWrapper.from_config` method. This method invokes `torch.load(..., weights_only=False)` without requiring an explicit unsafe opt-in, such as a `safe_mode=False` parameter. When called outside a `SafeModeScope(True)` context, the absence of an ambient safe mode state permits unsafe deserialization by default. This issue can lead to arbitrary code execution if untrusted Keras layer configurations are processed using this method. The vulnerability arises because the method does not enforce safe deserialization practices unless explicitly guarded by Keras safe mode.

CVSS3: 7.8
debian
13 дней назад

A vulnerability in keras-team/keras version 3.15.0 allows unsafe deser ...

CVSS3: 7.8
github
12 дней назад

A vulnerability in keras-team/keras version 3.15.0 allows unsafe deserialization of attacker-controlled PyTorch pickle data through the public `keras.layers.TorchModuleWrapper.from_config` method. This method invokes `torch.load(..., weights_only=False)` without requiring an explicit unsafe opt-in, such as a `safe_mode=False` parameter. When called outside a `SafeModeScope(True)` context, the absence of an ambient safe mode state permits unsafe deserialization by default. This issue can lead to arbitrary code execution if untrusted Keras layer configurations are processed using this method. The vulnerability arises because the method does not enforce safe deserialization practices unless explicitly guarded by Keras safe mode.

CVSS3: 7.8
fstec
3 месяца назад

Уязвимость функции TorchModuleWrapper.from_config() библиотеки Keras, позволяющая нарушителю выполнить произвольный код

EPSS

Процентиль: 11%
0.00204
Низкий

7.8 High

CVSS3