Hugging Face Packages Weaponized With a Single File Tweak

Hugging Face AI models can be manipulated by tweaking a single file in the tokenizer library, allowing hijacking of model outputs and exfiltration of data. This vulnerability poses a significant risk to users of Hugging Face models. The manipulation can be done by modifying a single file.