Hugging Face Malware Campaign Returns to Hijack GPUs
Original source
Hugging Face malware campaign returns to hijack GPUs for crypto mining
BleepingComputer →A cryptomining and backdoor campaign operating on Hugging Face has resumed operations days after takedown. Researchers at Pluto Security discovered 31 malicious repositories spread across seven accounts deploying models disguised as Qwen2 checkpoints that target self-hosted GPU inference servers. The attack exploits Hugging Face’s trust_remote_code feature, which executes model-supplied Python code automatically on load. The most popular malicious model, MENXIANGYU/qwen2-tiny, contained no actual model but installed a hidden cryptocurrency miner and reverse shell backdoor, accumulating more than 28,000 downloads.
The campaign’s persistence underscores the failure of existing security measures. After the first wave was removed by Hugging Face and ModelScope, 19 new malicious repositories appeared under different accounts within days. The flagship model’s downloads grew from 7,354 at reporting time to 28,412 by October 28, despite scanning clean through Hugging Face’s security partners. A replacement copy uploaded October 25 by account htt0118 remained live and functional.
Concurrently, researchers identified a GitHub repository posing as a Solana starter kit, created in November 2025, designed to trick AI coding agents into executing malware. The fake project’s only functional content is a PowerShell one-liner that downloads a script installing NetSupport Manager, a legitimate remote access tool configured for hidden attacker control. Security researchers linked the campaign to known NetSupport operators and previous schemes targeting cryptocurrency traders.
Read the full article
Continue reading at BleepingComputer →This is an AI-generated summary. Read the original for the full story.