The huggingface datasets library versions up to and including 5.00 contain a symlink-following vulnerability in the Extractor.extract() method. This vulnerability allows local attackers to pre-plant symbolic links at predictable output locations, causing the extraction process to follow these symlinks and write files to arbitrary filesystem paths. This behavior can be exploited in shared-cache environments to overwrite sensitive files, which may lead to privilege escalation or code execution. The vulnerability is addressed in a commit identified as ad2d853, but no official patch or vendor advisory details are provided in the input data.

Local attackers can exploit this vulnerability to overwrite arbitrary files on the filesystem by leveraging symbolic links during archive extraction. This can result in modification or replacement of sensitive files, potentially leading to privilege escalation or remote code execution in environments where shared caches are used.

Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. The vulnerability is reportedly fixed in commit ad2d853; therefore, upgrading to a version including this fix is recommended once available. Until then, avoid running extraction processes in shared-cache environments with untrusted archives or inputs.