What happened
finetrainers' Model Index Handler resolve_component_cls fails to validate attacker-influenced component names, allowing code injection. Assigned CVSS 3.1 6.3 / CVSS 4.0 2.1; CNA is cna@vuldb.com and the record is in Deferred status because the project has not responded.
Why it matters
A code-injection flaw in a Hugging Face fine-tuning library that can execute on the researcher's machine when a malicious or poisoned config is loaded — relevant to the supply-chain risk of pulling model/index/component definitions from the HF ecosystem.
Attack vector
Manipulating the 'Library' argument passed to resolve_component_cls in the Model Index Handler results in code injection; can be initiated remotely but requires user interaction, so it is typically triggered by loading an untrusted config/component
Affected systems
huggingface finetrainers 0.1.0 and 0.2.0 (finetrainers/utils/model.py, Model Index Handler)
Mitigation
No patch yet (project unresponsive); avoid loading untrusted finetrainers model-config/component inputs and pin/verify any components before use; monitor the repo for a fix (https://github.com/huggingface/finetrainers/)