Researchers hijack Hugging Face account for popular Kokoro AI model

Researchers took control of the Hugging Face account for Kokoro — a popular open-weight text-to-speech model with nearly 13 million monthly downloads, used in projects from Google, NVIDIA, AMD, Intel and AWS — by registering the developer’s lapsed domain and resetting the account password, showing how a hijacked publisher account could push poisoned model weights that no malware scanner would flag. They altered the model’s weights locally in 87 seconds so it spoke different words than the input text, without touching any code, and say Hugging Face did not respond to their report.

Source: www.bleepingcomputer.com

Curated Permalink