Fake ChatGPT and Claude Desktop Apps on GitHub Are Delivering Crypto Miners

Check Point Research has described a campaign dubbed Yoba, in which attackers pose as desktop apps for Anthropic Claude and OpenAI ChatGPT. They create fake GitHub projects with names such as GPT-5.6 Sol, GPT-6 Astra and Claude Opus, which deliver a malware family called YobaLoader. The attackers also set up GitHub.io pages describing the software's features to look more credible. The projects' popularity does not appear to be artificially boosted, but there are signs that SEO poisoning is used to push the links to the top of Google search results.

The malware is placed next to legitimate, signed executables from OpenAI and Anthropic so it can be loaded indirectly. This technique, known as DLL sideloading, is often used to bypass security protections. Each repository uses a different email address, and the uploader appears to operate in a different time zone each time, which Check Point says likely reflects a deliberate attempt to hide the attacker's real working hours.

The ZIP archives in these repositories contain a dropper called YobaDropper. It only runs when launched under the name Setup.exe, so it fails if analysed on its own in a sandbox. It also waits for a very long time, equal to about 12.73 days, before acting. It skips machines with keyboard layouts from Russia, Belarus, Tajikistan, Armenia, Azerbaijan, Georgia, Kazakhstan, Kyrgyzstan, Turkmenistan and Uzbekistan, and also skips the Tatar language setting. Ukraine is notably absent from that list.