Hackers exploit fake GitHub repos to mine cryptocurrency using trusted AI installer signatures, tricking users into downloading malicious files.
Malicious files mimic trusted AI software
Hackers have created fake installation files for OpenAI and Anthropic's AI models, embedding malicious code that can mine cryptocurrency on users' computers without their knowledge. These files are hosted on fake GitHub repositories and appear alongside legitimate files signed by the companies.Fake GitHub repos appear in search results
Google search results have shown some of these fake repositories at the top, possibly due to search engine poisoning. Check Point Research has been tracking this campaign, named Yoba, for two months and has identified how hackers trick users into downloading the malicious files.Malicious code disguises itself as a legitimate file
Hackers place a malicious file, such as a fake winmm.dll, next to a genuine installer signed by OpenAI or Anthropic. When users run the installer, the malicious file is loaded instead of the legitimate Windows multimedia library, allowing the hackers' code to run unnoticed.Attack evades detection and targets specific users
The hackers avoid computers using Russian or other CIS country keyboard layouts, except Ukraine, and target computers with French layouts. They also include long pauses in their code to avoid automated analysis tools, and Check Point has found indicators suggesting a Russian origin.Cryptocurrency mining uses victims' processors
Once a computer is infected, the hackers use a scheduled task disguised as a Microsoft task to maintain control. They execute their code through a hidden image on a Telegram channel and use Google's encrypted DNS service. The XMRig miner, connected to the Tor network, uses the victim's processor to mine cryptocurrency.Signs of covert mining include computer slowdowns
According to ESET, users may notice a slight slowdown, increased heat, or fan noise from their computer. On laptops, the battery might even swell and damage the device. Hackers can replace the miner with other malicious programs, such as password stealers, at any time.Other attacks use similar tactics
In March, hackers used fake pages for Claude Code to steal passwords and session cookies. In June, the Yoba group disguised their malware as a programming agent called Cursor. Users are advised to access ChatGPT or Claude only through the official websites of OpenAI and Anthropic.Users should check their systems and avoid untrusted sources
Check Point researchers warn companies to be cautious of AI assistants from unverified repositories. If a computer's fan is running loudly without activity, users should check the Task Manager for unusual CPU usage. Those who have installed applications from GitHub in the past few months are advised to run a full antivirus scan.
