Skip to content

Technology and money in plain English · Since 2012

Tech
News

Yoba Malware: Fake ChatGPT and Claude Desktop Apps on GitHub Install a Cryptominer

Check Point Research says fake GitHub projects posing as ChatGPT and Claude desktop apps sideload a loader that ends in a cryptominer. The legitimate signed files in the package are not the problem; the extra DLL beside them is.

Key facts

  • Check Point says lures were fake GitHub projects named for ChatGPT and Claude desktop apps As of
  • Final payload observed was the XMRig cryptominer As of
In this story5 sections
arda savasciogullari / Shutterstock.com
Listen to this article
5 min listenTech

The Yoba malware campaign uses fake GitHub projects that pose as ChatGPT and Claude desktop apps to install a cryptominer, according to a Check Point Research report published October 27. The unusual part is that the installers inside the downloads are real, signed files. The harm comes from a malicious DLL packaged beside them.

How the Yoba malware reaches victims

Check Point says the attackers built GitHub repositories and GitHub Pages sites with lure names such as “GPT-5.6 Sol,” “GPT-6 Astra” and “Claude Opus.” The download links pointed to ZIP files hosted as GitHub releases. Check Point reports that the projects ranked prominently in Google Search and assesses the campaign as consistent with SEO poisoning. It found no sign of artificially inflated stars or forks, so the pages did not look popular in the usual way.

Check Point also says it found no evidence that OpenAI, Anthropic, ChatGPT or Claude were compromised. The Yoba malware operators borrowed the real apps’ names and files, not their infrastructure.

Advertisement

Why a valid signature did not help

The ChatGPT-themed packages included a legitimate, signed OpenAI Setup.exe, and the Claude package included a legitimate, signed Claude executable. Each one loads a malicious winmm.dll that sits in the same folder, a technique called DLL sideloading. A code signature vouches for the signed executable only. It says nothing about other files packaged with it, which is the gap the Yoba malware exploits.

What the infection does

Check Point describes a multi-stage chain. The first stage, which it calls YobaDropper, checks that it was launched by the expected Setup.exe and delays if not. It applies regional language checks and requires administrator privileges. It then places a second component, YobaLoader, next to a legitimate Windows executable and creates a scheduled task so that executable loads YobaLoader after startup, which keeps the infection in place after a reboot.

YobaLoader finds its command server in an unusual way, according to the report. It reads the profile image of a Telegram channel and uses data from that image to generate a domain, so changing the image changes the domain. It then queries Google’s DNS-over-HTTPS service for a DNS TXT record on that domain, decrypts the response, and downloads an encrypted payload from external hosting such as Catbox.

The final payload Check Point observed was XMRig 6.26.0, a cryptominer, which communicated over Tor and paused mining when the user was active. That pause is meant to avoid the sluggishness that would prompt someone to investigate. Check Point notes that YobaLoader is a generic loader, so a different campaign could use it to deliver other Windows executables or PowerShell payloads. The blog post does not list specific hashes, domains or repository names and points to a full technical report for those.

What users and IT teams can do

Check Point’s advice is straightforward. Get AI software only from vendor-controlled channels, not from search results, GitHub pages or third-party links. Limit which applications can run and who has administrator rights, since the first stage needs elevated privileges. Inspect what a download actually contains and how it behaves rather than trusting the main file’s signature, and watch for suspicious network activity after installation, including generated domains.

For workplaces, the risk from Yoba malware and copycats grows as AI apps spread. A company that rolls out a tool widely, as Uber did with ChatGPT Work for its 30,000 employees, creates an obvious lure for anyone searching for the installer. And organizations that restrict AI tools, such as the Swedish government offices that blocked ChatGPT and Claude in the browser, may find employees hunting for workarounds that lead to exactly this kind of download.

Editor’s take

Treat the address, not the logo, as the product. A search result that looks right and a file that is signed can both be real and still lead to a bad install. For any AI desktop app, go to the vendor’s own site by typing it, or use the official app store, and ignore model names that sound like leaks or early access. The names Check Point lists are exactly the sort of thing people search for when they want the newest model first. Check Point’s report is the only source for the Yoba malware details so far, and the findings are best read as one security vendor’s analysis.

See something that needs fixing? Report an error

Get The Jolt Journal in your inbox

The day’s top stories in tech, finance and AI, picked by our editors. Free.

Choose your newsletters

Up next

Keep reading

Add The Jolt Journal as a preferred source on Google

Add The Jolt Journal to your Home Screen to get alerts

  • 64 Stories published

The Jolt Journal Tech Desk covers artificial intelligence, software, devices, cybersecurity and the platforms that shape how we live and work. Every story is checked against its sources and reviewed by our editor before publication.

Covers

Most Popular

More in Tech

See all

Leave a comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Breaking-news alerts

We’ll send a notification to this device when a big story breaks. Choose what you want, then allow notifications when your browser asks. You can change this or turn alerts off any time.

Send me