Cybercriminals are distributing the YobaLoader malware and cryptocurrency miners by masquerading as legitimate desktop applications for AI tools like OpenAI’s ChatGPT and Anthropic’s Claude on GitHub.
Key Points
- Attackers use SEO poisoning to surface fake GitHub projects, such as "GPT-6 Astra" and "Claude Opus," in search results.
- The malware employs DLL sideloading by pairing malicious files with legitimate, signed executables from OpenAI and Anthropic to bypass security.
- YobaDropper includes anti-analysis features, such as long sleep timers and geographic filters that exclude several Commonwealth of Independent States countries.
- The infection chain uses a domain generation algorithm and Google’s DNS-over-HTTPS to fetch payloads while evading network monitoring.
- Final payloads typically inject an XMRig cryptocurrency miner into the system process "svchost.exe" to hijack victim computing resources.