Skip to page content

AEGLIX MIND

Connected intelligence.Security research, connected

AEGLIX news · Global

Researchers Demonstrate Account Takeover of Popular Kokoro AI Model on Hugging Face

Based on reporting from BleepingComputer

Published by the source: Tue, 17 Nov 2026 13:21:24 -0500

AEGLIX News

Security researchers demonstrated they could hijack the Hugging Face account hosting the popular Kokoro text-to-speech model by exploiting a lack of ownership verification on the platform, according to a report published Tuesday. The team took control of the account by registering the hexgrad.com domain listed in the project's public DONATE.md file and using it to perform a password reset on the publisher's account. This vulnerability could allow an attacker to use the compromised account as a distribution channel for supply-chain attacks, potentially leading to data theft or code execution, the researchers warned.

Kokoro is an open-weight model that converts text into speech, supports multiple languages, and offers more than 50 voice presets. The model has nearly 13 million monthly downloads on Hugging Face, making it a high-value target for supply-chain compromise. The researchers' takeover method relied solely on the domain registration and password reset process, highlighting a systemic gap in how the platform verifies model ownership before granting account control.

In a controlled test, the researchers modified Kokoro's model weights in 87 seconds using a gaming GPU, causing the model to produce audio that differed from the input text without altering the underlying application code. Because only the weights were changed, the poisoned output would leave no malicious code or signature for malware scanners to detect. Tests from a separate account confirmed that Hugging Face would not flag the modified release as suspicious, and the release would contain no detectable malware, demonstrating the stealth of such a weight-based supply-chain attack.

The researchers reported the issue to Hugging Face but received no response, according to the team. BleepingComputer also contacted Hugging Face for comment but had not received a response at the time of publication. The researchers emphasized they never uploaded the modified weights to Hugging Face during their testing, limiting the demonstration to a local environment.

Report prepared 2026-11-17 · Based on reporting from BleepingComputer. Source publication date above; subsequent developments may not be reflected.

Sources

Read original source · BleepingComputer (opens in a new tab)

Privacy policy

Explore the story

Consult the cited publisher for the original reporting, updates and supporting evidence. Check the publication date and source before acting on a headline.

These are research tools; no CVE, indicator, or campaign relationship is inferred from this article.

More from BleepingComputer

← Back to security news