Threat

Researchers hijack Hugging Face account for popular Kokoro AI model

Page published

Earliest dated coverage: 17 Nov 2026 · First observed: 17 Nov 2026 · Latest dated coverage: 17 Nov 2026

Coverage timeline

17 Nov 2026bleepingcomputer.com

Single-source incident — one report is available.

Why it matters

The Kokoro account takeover demonstrates how abandoned publisher domains can expose widely used AI models to supply-chain tampering, including weight changes that alter behavior without introducing malicious application code.

Researchers took control of the Hugging Face publisher account behind Kokoro, a text-to-speech model with nearly 13 million monthly downloads, by registering the original owner's domain and resetting the account password. In a separate local test, they modified model weights in 87 seconds to make the generated speech differ from the input text without changing application code. The researchers said they never uploaded the modified weights to Hugging Face.