Critical Bugs Put Hugging Face AI Platform in a 'Pickle'

Wiz researchers discovered and demonstrated two critical security flaws in Hugging Face's inference stack: allowing Pickle-based models to execute arbitrary code (enabling reverse shells and exploration of the EKS-hosted environment that could lead to cross-tenant access to secrets) and the ability during Spaces build-time to access and potentially overwrite images in a shared container registry—raising supply-chain and model/data integrity risks; Hugging Face says it has mitigated the issues.