Fake OpenAI Privacy Filter Repository Tops Hugging Face Trending to Distribute Rust-Based Info Stealer

A malicious repository on the Hugging Face platform impersonating OpenAI's Privacy Filter open-weight model ascended to the top of the trending list, attracting over 244,000 downloads. This fake project deployed a Rust-based information stealer targeting Windows users, highlighting risks associated with impersonated open-source models.

What happened

A repository named Open-OSS/privacy-filter appeared on Hugging Face, closely mimicking OpenAI's official Privacy Filter model, which was legitimately released recently under openai/privacy-filter. The malicious repo copied the entire codebase and branding to create an authentic appearance. This strategy facilitated widespread downloads and trust among users. However, the repository contained a Rust-based information-stealing component designed to target Windows platforms, posing significant security threats to anyone who executed or integrated the compromised model.

Why it matters

This incident underscores the vulnerability of open-source and model-sharing platforms to malicious actors who can exploit user trust through repository impersonation. Developers and organizations relying on openly shared AI models must exercise caution and verify repository authenticity before integrating code. The presence of a sophisticated Rust-based info stealer embedded in what appeared to be a legitimate project demonstrates how attackers use popular platforms to distribute malware, increasing the difficulty of detecting threats in rapidly evolving AI ecosystems.

What security teams should do

Security teams should verify the source and authenticity of machine learning models and code repositories prior to usage. Monitoring downloads and scan reports on platforms like Hugging Face can help identify suspicious activity. It is essential to audit any recently downloaded or integrated repositories for unusual binaries or scripts, particularly when sourced from trending but unofficial repositories. Endpoint detection solutions should be deployed to identify behaviors associated with information stealers on Windows systems. Security teams may also wish to review any environments where this repository or model has been used to assess impact and prevent data leakage.

Key technical details

The malicious repository, Open-OSS/privacy-filter, copied the entire codebase of OpenAI's legitimate Privacy Filter open-weight model to appear genuine. It delivered a Rust-based information stealer specifically targeting Windows operating systems. This indicates a deliberate attempt to embed advanced native malware within what ostensibly looks like AI-related open-source software. Details such as the Rust component’s specific capabilities, communication methods, or persistence techniques were not disclosed in the source material. The original legitimate model was released by OpenAI late last month under openai/privacy-filter.

Affected organizations/products

The malicious repository affected users downloading or using the fake Open-OSS/privacy-filter project from Hugging Face, primarily impacting Windows users due to the targeting of a Rust-based info stealer payload. The legitimate model by OpenAI remains unaffected in terms of platform status, but users may be confused by the impersonation.

Source attribution

https://thehackernews.com/2026/05/fake-openai-privacy-filter-repo-hits-1.html