Safety · 2026-10-30
Fake AI Models on Hugging Face Mined Cryptocurrency on Users' Computers
Dozens of fake AI models on Hugging Face were found secretly running cryptocurrency mining on victims' computers, and one of them was downloaded more than 28,000 times.
What happened
Researchers found 31 malicious repositories on Hugging Face, the platform where developers share AI models. They were uploaded through 7 accounts and posed as small test versions of the Qwen2 model.
Before October 28, the most popular malicious model had been downloaded 28,412 times. The total number of downloads for all related models exceeded 32,000.
The discovery was reported by BleepingComputer, citing researchers at Pluto Security.
How it worked
The malware was built for Linux, an operating system often used to run AI models on personal computers.
When a developer downloaded such a model and allowed it to run its code, a cryptocurrency miner was installed on the computer along with it. The program disguised itself as a normal system process and tried to restart after being stopped.
Some versions were even more dangerous. They allowed the hacker to connect to an infected computer over the internet and run commands without the owner's knowledge.
Why security checks did not stop it
Hugging Face scans uploaded files with antivirus software and specialized tools. However, these mainly look for threats that are already known, so new malicious code could go unnoticed.
Hugging Face has already blocked the most popular malicious model. However, at the time of BleepingComputer's report, an almost identical copy was still available.
What to do if you downloaded such models
Check your computer for infection. Removing the miner alone may not be enough, because the hacker could have left programs that let them get back into the system.
Change passwords and access keys for the services you use on this computer.
If you cannot fully remove the infection, you may need to reinstall the operating system.
General rule: download models only from trusted sources and be careful about running code from others.
Useful from the source
- BleepingComputer: article on malicious AI models on Hugging Face — English-language article that is the source of the main facts about the campaign.
- Pluto Security: research on the campaign — Blog post from Pluto Security, the company that analyzed the malicious models.
Sources: 🔗 DOU/dev.ua