Hugging Face repositories hijacked to deploy cryptominers and backdoors

A cryptomining and backdoor campaign abusing Hugging Face model repositories has returned days after being taken down, BleepingComputer reported. Researchers at Pluto Security discovered 31 malicious repositories spread across seven Hugging Face accounts, which were published in two separate waves starting in September. The operators used these poisoned models to hijack server graphics processing units for cryptocurrency mining and to plant persistent backdoors. [1]

The most popular malicious model in the campaign was downloaded more than 28,000 times before detection, according to BleepingComputer. In an unrelated incident highlighted by researchers, a fake Solana starter kit hosted on GitHub since late 2025 was designed to trick artificial intelligence coding agents into installing a NetSupport remote access trojan. [1]

Hugging Face is an American platform headquartered in New York City that enables users to host and share datasets and machine learning models. Its central Hub hosts over 3 million models, 500,000 datasets, and 1 million applications that serve millions of developers. [3][5]

Software supply chain attacks target less secure upstream components, such as open source libraries and model repositories, to compromise downstream corporate networks. Organizations using public machine learning repositories must verify model provenance and inspect code dependencies before running third party artifacts in production environments. [1][4]