Threat actors compromised the PyTorch Machine Learning Framework by adding a malicious dependency.
The maintainers of the PyTorch package warn of a supply chain attack. According to an advisory published by the maintainers, users who have installed PyTorch-nightly on Linux via pip between December 25, 2022 and December 30, 2022, must uninstall it and use the latest binaries.
PyTorch is a machine learning framework based on the Torch library, used for applications such as computer vision and natural language processing, originally developed by Meta AI and now part of the Linux Foundation umbrella.
PyTorch, analogous to Keras and TensorFlow, is an open source Python-based machine learning framework that was originally developed by Meta Platforms.
The attack was discovered on December 30. The threat actors uploaded a malicious version of a legitimate dependency named torchtriton to the Python Package Index (PyPI) code repository.
The supply chain attack directly affects dependencies for packages that are hosted on public package indices.
In order to determine if a Python environment is affected or not, the maintainers recommends to run the following command to searches for the malicious binary in the torchtriton package (PYTHON_SITE_PACKAGES/triton/runtime/triton):
python3 -c “import pathlib;import importlib.util;s=importlib.util.find_spec(‘triton’); affected=any(x.name ==
The advisory points out that the malicious binary is executed when the triton package is imported, which requires explicit code to do.
Bleeping Computer reported that the owner of the domain to which the stolen data was transmitted claimed it was part of an ethical research exercise and that all the data has since been deleted.
As mitigations, torchtriton has been removed as a dependency and replaced with pytorch-triton. A dummy package has also been registered on PyPI as a placeholder to prevent further abuse.
“This is not the real torchtriton package but uploaded here to discover dependency confusion vulnerabilities,” stated the PyPI page for torchtriton, now removed. “You can get the real torchtriton from https://download.pytorch[.]org/whl/nightly/torchtriton/.”
Comments