Fake AI Repo Infects 240K+ PCs

A sophisticated cyberattack recently saw attackers successfully distribute an infostealer malware by creating a fake OpenAI repository on HuggingFace. This malicious "privacy filter" model, disguised as a legitimate tool, garnered an alarming 244,000 downloads before its discovery and removal.

The malware employed a multi-stage attack, first disabling SSL verification and then executing a PowerShell command to escalate privileges. It successfully added itself to Microsoft Defender's exclusion list, allowing the "sefirah" payload to deploy and steal a wide range of sensitive information. This included browser data, Discord tokens, crypto wallet details, and various credentials.

While the download numbers are high, some indicate potential inflation due to auto-generated "likes." Nevertheless, the fake repository briefly topped HuggingFace rankings, demonstrating its potential reach. Security researchers tracked down and exposed other linked malicious repositories, all of which have since been removed from the platform, mitigating further risk.