Return to Threats

Researchers hijack Hugging Face account for popular Kokoro AI model

BleepingComputer 2026-11-17 AI supply chain Critical

Executive Summary

Researchers showed that compromising the account distributing the Kokoro AI model enabled modification of its weights and changed model output. Organizations using AI models or repositories should treat model distribution channels as part of their software supply chain.

What Happened

Researchers demonstrated takeover of the account distributing the Kokoro AI model and modified its weights in a controlled test. The altered model produced audio different from the input text, showing how compromised model repositories could distribute stealthy poisoned artifacts.

Affected Products

  • Kokoro AI model
  • Hugging Face

Vulnerability / Attack Technique

AI model repository account takeover and model poisoning

Technical Mechanism

The researchers took over the account distributing the Kokoro AI model in a controlled test and modified the model weights. The altered weights caused the model to produce audio that differed from the input text.

AI Security Category

LLM04:2025 Data and Model Poisoning

Part of the AI Supply Chain and AI SBOM topic.

Relevant Frameworks

LLM03:2025 Supply ChainLLM04:2025 Data and Model PoisoningMITRE ATLAS: Initial AccessMITRE ATLAS: Impact

RealGround Analysis

Compromised model repositories can undermine the integrity of models before they reach AI agents or LLM applications. Teams should maintain an inventory of model dependencies and test downloaded artifacts for integrity and unexpected behavior before deployment.

Recommended Controls

  • Track all external models and model repositories in an AI asset and dependency inventory.
  • Verify model provenance and cryptographic integrity before deployment and after updates.
  • Require review and approval for model-weight changes or new model versions.
  • Run behavioral and security tests to detect outputs that diverge from expected inputs.
  • Restrict repository-account privileges and protect them with strong authentication and monitoring.

Related RealGround Service

AI Agent SBOM and LLM Mapping · or start with the free AI Agent Security Scanner.

Sources

BleepingComputer: https://www.bleepingcomputer.com/news/security/researchers-hijack-hugging-face-account-for-popular-kokoro-ai-model/

Summary and analysis written by RealGround from the source above; last updated 2026-11-20.

Talk to AI CISO