Malicious Software Packages Dataset

This repository is an open-source dataset of 3478 malicious software packages (and counting) identified by Datadog, as part of our security research efforts in software supply-chain security. Most of the malicious packages have been identified by GuardDog.

Current ecosystems:

• PyPI
• NPM

Usage

Malicious samples are available under the samples/ folder and compressed as an encrypted ZIP file with the password infected. The date indicated as part of the file name is the discovery date, not necessarily the package publication date.

You can use the script extract.sh to automatically extract all the samples to perform local analysis on them. Alternatively, you can extract a single sample using:

$ unzip -o -P infected samples/pypi/pydefender/1.0.0/2023-03-20-pydefender-v1.0.0.zip -d /tmp/
Archive:  samples/pypi/pydefender/1.0.0/2023-03-20-pydefender-v1.0.0.zip
   creating: /tmp/2023-03-20-pydefender-v1.0.0/

Each samples/ subdirectory contains a manifest.json file that identifies the packages, and the versions of those packages, that comprise the samples collected for each ecosystem. You can use these files to quickly search the dataset for particular samples.

License

This dataset is released under the Apache-2.0 license. You're welcome to use it with attribution.

You can cite it using:

@misc{OpenSourceDatasetMaliciousSoftwarePackages, 
    month     = Mar,
    day       = 20,
    date      = 2023,
    journal   = {Open-Source Dataset of Malicious Software Packages},
    publisher = {Datadog Security Labs},
    url       = https://github.com/datadog/malicious-software-packages-dataset, 
}

Malicious software packages provided as part of this repository may contain legitimate, licensed code. In that case, the applicable license is the one of the original package, indicated in the metadata of its setup.py file.

Disclaimers

• This repository contains actively malicious software that was published by threat actors. Do not run it on your machine.

• This dataset may suffer from selection biais, as it was mostly identified by a single ruleset (GuardDog). As such, it may not accurately represent the landscape of software supply-chain security malware.

• Some of the packages in this dataset are very similar. As of March 20st 2023, we estimate that the dataset contains around 50 clusters of highly similar packages. One of the clusters is very large (more than 500 packages highly similar to this one), which might skew the dataset.

FAQ

Are you maintaining this dataset?

We will be regularly adding new packages to the dataset.

How do you know these packages are malicious?

Every single software package included in this dataset has been manually triaged by a human.

What if the manifest.json entry for a package has an empty version list?

Around 250 packages in the PyPI subset do not have any affected versions listed in their manifest.json entries. These cases are holdovers from the earliest days of the project before version information was attached to the sample names.

If you intend to use this dataset to screen packages for known-maliciousness, then all versions of packages with empty version lists should be considered malicious.

How are you clustering these packages?

At the time, we did not make available the clustering algorithm we use internally to group similar samples and ease analysis. If you have interest, please reach out at [email protected] - we'll be happy to talk!

Do you accept contributions?

At the time, the repository is not accepting contributions. However, if you'd like to share an interesting finding with us, reach out at [email protected]!

Other datasets

https://github.com/lxyeternal/pypi_malregistry and related paper https://lcwj3.github.io/img_cs/pdf/An%20Empirical%20Study%20of%20Malicious%20Code%20In%20PyPI%20Ecosystem.pdf https://github.com/cybertier/Backstabbers-Knife-Collection