AI news ·
Hugging Face Malware Campaign Returns After Takedown
Researchers found 31 Hugging Face model repos with miners and backdoors, downloaded over 32,000 times, showing the same threat returned after earlier removal.
Researchers identified 31 malicious model repositories on Hugging Face that were collectively downloaded more than 32,000 times. The repositories contained cryptocurrency miners and backdoors, marking a return of the same threat cluster after a prior enforcement action removed earlier versions of the campaign.
The findings highlight that model repositories and the remote code they can execute must be treated as software supply-chain inputs. Behavioral inspection of models and their dependencies is necessary to detect payloads that static analysis alone may miss.
For operations and development teams working with AI model registries, the incident reinforces the need to apply supply-chain security practices to machine learning artifacts. Treating model files as untrusted code and subjecting them to runtime behavioral checks can reduce the risk of reintroducing threats that reappear after takedowns.