Описание
A vulnerability in the LightGlue model loading path of huggingface/transformers version 5.2.0 allows an attacker-controlled model repository to execute arbitrary code during model initialization. The issue arises because the trust_remote_code parameter, intended to prevent remote code execution, is overridden by untrusted serialized configuration data in a nested code path. Specifically, when loading a LightGlue model using AutoModel.from_pretrained() with trust_remote_code=False, the LightGlueConfig reads the trust_remote_code value from the untrusted config.json file and propagates it into nested AutoConfig.from_pretrained() calls. This results in the execution of attacker-provided Python modules, even when the victim explicitly disables remote code execution. The vulnerability poses a high risk for environments such as API inference servers, research notebooks, CI/CD pipelines, and model evaluation workers, potentially leading to credential theft, lateral movement, or persistence/backdoor deployment.
A flaw was found in python-transformers. An attacker can exploit this vulnerability by providing a malicious model repository. During model initialization, the trust_remote_code parameter, intended to prevent remote code execution, is overridden by untrusted configuration data. This allows the attacker to execute arbitrary code, potentially leading to credential theft, unauthorized access, or the deployment of backdoors.
Отчет
This Important vulnerability in python-transformers allows for arbitrary code execution when loading a malicious LightGlue model, even if trust_remote_code is explicitly set to False. This occurs because untrusted serialized configuration data can override the security setting in nested model loading paths. Red Hat products utilizing python-transformers for AI/ML workloads, such as Red Hat AI Inference Server and Red Hat OpenShift AI, are affected, posing a significant risk of credential theft or system compromise if untrusted models are processed. The LightGlue model (and its vulnerable trust_remote_code handling) was introduced in transformers 4.53.0 and fixed in 5.5.0; the CNA-published affected range is all versions before 5.5.0.
Меры по смягчению последствий
Mitigation for this issue is either not available or the currently available options do not meet the Red Hat Product Security criteria comprising ease of use and deployment, applicability to widespread installation base or stability.
Затронутые пакеты
| Платформа | Пакет | Состояние | Рекомендация | Релиз |
|---|---|---|---|---|
| OpenShift Lightspeed | openshift-lightspeed/lightspeed-ocp-rag-rhel9 | Not affected | ||
| OpenShift Lightspeed | openshift-lightspeed/lightspeed-service-api-rhel9 | Not affected | ||
| OpenShift Lightspeed | openshift-lightspeed-tech-preview/lightspeed-rag-tool-rhel9 | Not affected | ||
| Red Hat AI Inference Server | rhaii/model-opt-cuda-rhel9 | Affected | ||
| Red Hat AI Inference Server | rhaiis/model-opt-cuda-rhel9 | Affected | ||
| Red Hat AI Inference Server | rhaiis/vllm-cpu-rhel9 | Will not fix | ||
| Red Hat AI Inference Server | rhaiis/vllm-cuda-rhel9 | Affected | ||
| Red Hat AI Inference Server | rhaiis/vllm-neuron-rhel9 | Will not fix | ||
| Red Hat AI Inference Server | rhaiis/vllm-rocm-rhel9 | Affected | ||
| Red Hat AI Inference Server | rhaiis/vllm-spyre-rhel9 | Affected |
Показывать по
Дополнительная информация
Статус:
EPSS
7.7 High
CVSS3
Связанные уязвимости
A vulnerability in the LightGlue model loading path of huggingface/transformers version 5.2.0 allows an attacker-controlled model repository to execute arbitrary code during model initialization. The issue arises because the `trust_remote_code` parameter, intended to prevent remote code execution, is overridden by untrusted serialized configuration data in a nested code path. Specifically, when loading a LightGlue model using `AutoModel.from_pretrained()` with `trust_remote_code=False`, the `LightGlueConfig` reads the `trust_remote_code` value from the untrusted `config.json` file and propagates it into nested `AutoConfig.from_pretrained()` calls. This results in the execution of attacker-provided Python modules, even when the victim explicitly disables remote code execution. The vulnerability poses a high risk for environments such as API inference servers, research notebooks, CI/CD pipelines, and model evaluation workers, potentially leading to credential theft, lateral movement, or
A vulnerability in the LightGlue model loading path of huggingface/transformers version 5.2.0 allows an attacker-controlled model repository to execute arbitrary code during model initialization. The issue arises because the `trust_remote_code` parameter, intended to prevent remote code execution, is overridden by untrusted serialized configuration data in a nested code path. Specifically, when loading a LightGlue model using `AutoModel.from_pretrained()` with `trust_remote_code=False`, the `LightGlueConfig` reads the `trust_remote_code` value from the untrusted `config.json` file and propagates it into nested `AutoConfig.from_pretrained()` calls. This results in the execution of attacker-provided Python modules, even when the victim explicitly disables remote code execution. The vulnerability poses a high risk for environments such as API inference servers, research notebooks, CI/CD pipelines, and model evaluation workers, potentially leading to credential theft, lateral movement, ...
EPSS
7.7 High
CVSS3