Copy ChatGPT and Claude apps install malware with a good signature

Check Point Research tells that attackers use copy ChatGPT and Claude desktop apps for Windows to install malware. The campaign, with the name Yoba, uses copy GitHub pages that are high in Google results. Each download holds a Setup.exe with the signature of OpenAI or Anthropic, and a bad file that it starts. The bad releases were uploaded on 30 and 31 August and on 4 September.

How it works

The file with the signature starts a bad library file, winmm.dll. This is DLL sideloading: a good app starts a bad library. The bad file waits about 12 days and must have administrator access. It then makes a scheduled task that shows the name Microsoft.

What was seen

Check Point saw the XMRig miner as the last payload. A miner is software that makes cryptocurrency. It uses Tor and stops when the user is active. The loader can also get other Windows software. Check Point tells that signs point to Russia, but it did not name a group.

What is not known

OpenAI and Anthropic gave no comment on requests. Yash Kadakia of Security Brigade tells that each connection can look correct when a person checks only one connection. Kadakia tells firms to make approved AI tools easy to get.

Insight: Copy AI apps win through trust and high Google results

The Yoba pages did not add stars or forks, but they were in the top Google results, Check Point tells. On 9 October, a Google ad sent users to a copy Claude page. On 23 October, an infostealer was found that copies the logins of 7 or more AI tools. This reads as a method: attackers use known AI names as the first step, and Google results are how they find users.