# SZB-a5a96cf48f: Researchers take over Kokoro's Hugging Face account and alter its model weights

Security researchers say they gained control of the Kokoro text-to-speech model's Hugging Face account by re-registering the original owner's expired domain and resetting the password.

- Type: Breach
- Status: Corroborated
- Vendor: Hugging Face
- Product: Hugging Face platform
- Disclosed: 2026-11-17 18:21 UTC
- Sources: 2 distinct
- First seen by cve.now: 2026-11-17 18:22 UTC · Updated: 2026-11-17 18:25 UTC

> No statement or filing from Hugging Face has reached cve.now yet. Every claim below is quoted from, and linked to, the source that made it.

Written by a model from the sources below. Verify against them.

## Summary

They say they altered the model weights in a controlled test but did not upload the changes, and that Hugging Face did not respond to their report.

## What to do

- Pin model downloads to a verified commit hash instead of tracking the latest revision
- Check model weights against known-good checksums before deploying updates
- Review pipelines that auto-fetch Kokoro updates from Hugging Face
- Watch for unexpected publisher changes on models your systems depend on

## Timeline

- 2026-11-17 18:21 UTC: Made public by BleepingComputer
- 2026-11-17 18:22 UTC: First seen by cve.now
- 2026-11-17 18:23 UTC: [Posted on X by cve.now](https://x.com/i/status/2122757627020988699)

## Evidence (2 sources)

- 2026-11-17 · Press · [BleepingComputer](https://www.bleepingcomputer.com/news/security/researchers-hijack-hugging-face-account-for-popular-kokoro-ai-model/): “Hertz and his team took control of the Kokoro account on Hugging Face by simply registering the domain used by the original owner and resetting the login password.”
- 2026-11-17 · Press · [BleepingComputer on Bluesky](https://bsky.app/profile/bleepingcomputer.com/post/3n2jnvqhw2x2q): “A Lack of ownership verification on the Hugging Face platform allowed researchers to take control of Kokoro, a text-to-speech model used in projects from Google, NVIDIA, AMD, Intel, and Amazon Web Services.”

## Corrections

None so far.

Something wrong on this page? Report an error: https://github.com/sazabi/cve.now/issues/new?title=Wrong%20fact%20on%20SZB-a5a96cf48f&body=https%3A%2F%2Fcve.now%2FSZB-a5a96cf48f%0A%0AWhat%20is%20wrong%2C%20and%20the%20source%20that%20shows%20it%3A%0A

## Questions about the Hugging Face breach

### Was Hugging Face breached?

BleepingComputer reports it. No statement from Hugging Face has reached cve.now yet.

### What should Hugging Face customers do?

Pin model downloads to a verified commit hash instead of tracking the latest revision. Check model weights against known-good checksums before deploying updates. Review pipelines that auto-fetch Kokoro updates from Hugging Face. Watch for unexpected publisher changes on models your systems depend on.

## Related

- Started as early warning SZB-a5a96cf48f, first seen 2026-11-17 18:22 UTC
- Browse: [Breaches](https://cve.now/breaches.md) · [Early warnings](https://cve.now/early-warnings.md)

---

Source: cve.now
Built by [Sazabi](https://www.sazabi.com/?utm_source=cve-now&utm_medium=markdown), AI-native observability and security: automatic issue detection, plain-language log search and live dashboards.
Page: https://cve.now/SZB-a5a96cf48f
Site guide for agents: https://cve.now/llms.txt
