Researchers hijack Hugging Face account for Kokoro AI model
AI-generated summary Security researcher Alon Hertz and his team gained control of the Hugging Face publisher account for the Kokoro text-to-speech model by registering an abandoned domain listed in the project files and resetting the password. The team demonstrated that fine-tuning the model weights in a short test allowed the modified version to produce audio diverging from the input text while bypassing malware scanners since no code was altered.