Fake ChatGPT and Claude Desktop Apps on GitHub Spread Crypto Miners, Check Point Says

Security researchers at Check Point Research have flagged a campaign they call Yoba, in which fake GitHub projects posing as desktop apps for OpenAI's ChatGPT and Anthropic's Claude are used to deliver malware and cryptocurrency miners.

The repositories advertise names such as GPT-5.6 Sol, GPT-6 Astra and Claude Opus, and come with GitHub.io pages describing the software's features to appear legitimate. Check Point says the projects do not rely on ghost networks to inflate their popularity; instead, signs point to SEO poisoning that pushes their links to the top of Google results.

Each ZIP archive holds a legitimate signed OpenAI Codex or Claude executable, named Setup.exe, alongside a malicious winmm.dll. The DLL is sideloaded by the signed program, a technique used to slip past security protections. The dropper, YobaDropper, only runs when launched under Setup.exe, and it includes anti-analysis tricks: 200,000 sleep cycles of 5,500 milliseconds each, roughly 12.73 days, and checks that skip machines with keyboard layouts for Russia, Belarus, Tajikistan, Armenia, Azerbaijan, Georgia, Kazakhstan, Kyrgyzstan, Turkmenistan and Uzbekistan, plus the Tatar language ID. Ukraine, historically on such lists, is absent.

YobaDropper then triggers a second sideloading chain using wmpdmc.exe and ampdui.dll, drops a payload as wmpdui.dll and sets up persistence through a scheduled task. The next stage, YobaLoader, pulls an image from a Telegram channel, uses its last 50 bytes to generate a domain via a domain generation algorithm, and queries Google's DNS-over-HTTPS API for a Base64-encoded command. The fetched payload is injected into svchost.exe, running an XMRig cryptocurrency miner.

Check Point says the first repository appeared around July 2026, with malicious releases uploaded at the end of August. Earlier YobaLoader versions excluded only Russia and carried extensive anti-debugging and anti-VM checks that were later removed. The researchers note that packaging code with well-known software names and hosting it on GitHub lowers the odds that users or basic security checks will flag the files as suspicious.