Malicious Repository Impersonating OpenAI Models Ranks No. 1 on AI Sharing Platform

A malicious repository mimicking popular Artificial Intelligence (AI) models from OpenAI has topped the list of most-used programs on 'Hugging Face,' a popular site for sharing AI models. Security experts first discovered the issue on March 7th. The repository appeared to be a legitimate AI model but was confirmed to have been used to spread malware that secretly steals personal information stored on computers.

Before being disabled, the repository recorded 244,000 downloads and 667 likes over approximately 18 hours. Following the repository's instructions and executing specific files triggered a malicious script. This script then downloaded additional files from an external server, ultimately executing information-stealing malware. This malware was designed to collect sensitive personal information, including user screen captures, browser data, Discord account details, and cryptocurrency wallet-related data.

Hugging Face has currently blocked access to the malicious repository, and during the investigation, additional repositories distributing malware in a similar manner were discovered. These also used the tactic of impersonating famous AI models to trick users.

IT security news outlet The Hacker News pointed out that this case highlights the difficulty of judging the reliability of projects on AI open-source platforms based solely on popularity rankings or download counts. The ability for repositories impersonating well-known companies or models to quickly climb the popularity charts requires caution from developers and researchers.