AI Stack Alert Medium TD-2026-0115 For AI Eng, AppSec, CISO Unconfirmed AI-assisted
Researchers hijack abandoned Hugging Face account for Kokoro text-to-speech model
Researchers took control of the Hugging Face account for Kokoro, an open-weight text-to-speech model, by re-registering the domain listed in the developer's DONATE.md file and resetting the password. They modified the model weights in a controlled test so the output differed from the input text, without changing application code. They did not upload the modified weights, and Hugging Face did not respond to the report.
Why it matters
Organizations that automatically fetch new model revisions could receive altered weights through a routine update that looks legitimate and contains no malware signature. Kokoro is used in tools from Google, NVIDIA, Intel, AMD, and AWS guides.
Do this
- Inventory systems that use Kokoro or pull it from Hugging Face, PyPI, or GitHub, and record the pinned revision or hash for each.
- Pin model downloads to a specific commit hash and verify file checksums before deployment.
- Disable automatic model revision updates for production inference pipelines until publisher ownership is verified.
- Monitor model outputs for unexpected content changes, such as speech that does not match input text.
Signals
- NoCISA KEV
- 26/100Priority
- 70%Confidence
Evidence
“Hugging Face does not vet publishers. It doesn’t verify whether an account is still held by its original, legitimate owner”
“Because only the weights change, there is no malicious code or signature for malware scanners to detect, the researchers say.”
Sources
Affected / related
Kokoro, Hugging Face, Hexgrad
Published 2026-11-17 18:21 UTC · updated 2026-11-17 19:05 UTC · AI-assisted summary, checked against quoted sources · Unconfirmed: single non-official source · How we score