Have you ever downloaded a “free desktop version” of ChatGPT or Claude from a professional-looking GitHub project simply because it appeared at the very top of your search results? If your answer is yes, you may have invited an uninvited guest right into your computer.
Throughout 2026, a campaign named Yoba has silently exploited users' curiosity and desire to experience AI to distribute malware. The scariest part is not the suspicious files themselves, but the fact that the attackers have borrowed legitimate, digitally signed executable files from OpenAI and Anthropic to fool both operating systems and humans alike.
Let me, SilverZ, dissect this trick from its roots. Because when malware is no longer “ugly” but instead wears the cloak of prestigious AI brands, standard vigilance is simply not enough.
The Yoba Campaign: When Fake GitHub Projects Become Malware Distribution Hubs
The story begins with seemingly legitimate GitHub projects. Attackers create repositories with names like GPT-5.6 Sol, GPT-6 Astra, and Claude Opus—names carefully designed to appeal to the "newest and most powerful" mindset. Accompanying these are GitHub.io pages detailing software "features," creating a false impression of an active community, documentation, and sponsorship.
To increase credibility, the attackers even use different email addresses for each repository, with signs indicating that the uploaders operate in various time zones such as UTC+4, UTC-4, and UTC+3. This is a deliberate effort to obscure the attackers' true operational time zones.
And to ensure these fake projects appear right at the top of search results, the campaign relies on SEO poisoning techniques. In other words, the attackers don't just build a perfect cover; they actively push that cover right in front of the victims.
The “Parasitic” Technique: DLL Sideloading and Signed Files
This is the most critical technical aspect and the main reason why this campaign is so difficult for standard security solutions to detect.
Inside each distributed ZIP file are two main components:
- Setup.exe: A legitimate executable file digitally signed by OpenAI (OpenAI Codex) or Anthropic (Claude).
- winmm.dll: A malicious DLL file placed in the same directory as the legitimate executable.
When the user runs Setup.exe, this legitimate executable automatically loads winmm.dll—a technique known as DLL sideloading. Because Setup.exe is digitally signed and completely legitimate, operating systems and antivirus software typically raise no suspicions. However, the accompanying DLL file is far from harmless.
winmm.dll is actually a dropper named YobaDropper. It performs a series of checks to ensure it only runs when executed in the correct context:
- Checksum check: It hashes the name of the host executable and compares it against a hard-coded value of 0xE6C2F326. In other words, it only functions when executed under the name “Setup.exe” and will fail if run independently in a sandbox environment.
- 12.73-day delay: It performs 200,000 iterations of “Sleep(5500)”, which translates to waiting over 12 days before taking action. This is an anti-analysis technique designed to trick automated sandbox systems that typically run tests for only a short duration.
- Keyboard layout exclusions: It skips machines with keyboard layouts corresponding to Russia, Belarus, Tajikistan, Armenia, Azerbaijan, Georgia, Kazakhstan, Kyrgyzstan, Turkmenistan, and Uzbekistan. Notably, Ukraine is no longer on this list, marking a change from previous campaigns.
YobaDropper then triggers a second sideloading chain using wmpdmc.exe (copied from the system32 directory) and ampdui.dll, placing them into a new folder named “WMProcess” within ProgramData. This multi-layered attack chain is designed to maintain persistence while remaining stealthy.
The Ultimate Goal: YobaLoader and Crypto Miners
After bypassing all defense layers, the final downloaded malware is YobaLoader—a loader capable of fetching and executing additional payloads. In this campaign, the observed payloads are crypto miners.
The consequence is not just the hijacking of the victim's computer resources to mine cryptocurrency for the attacker. It also brings reduced system performance, higher electricity bills, and more importantly, a potential backdoor allowing attackers to return and deploy even more dangerous payloads in the future, ranging from data theft to ransomware.
Why Is This Trick Especially Dangerous?
What makes the Yoba campaign alarming is not just its sophisticated technique, but its combination of psychological and technical factors.
First, it preys on brand trust. When users see a Setup.exe file signed by OpenAI or Anthropic, they tend to trust it. But a digital signature only confirms that the file was created by OpenAI or Anthropic; it does not guarantee that the entire software package—including accompanying DLL files—is safe.
Second, it exploits a lack of understanding regarding Windows' dynamic library loading mechanism. Many users do not know that an EXE file can be “tricked” into loading a malicious DLL if that DLL is located in the same directory. This is an inherent design flaw in Windows that has been exploited for decades.
Third, it weaponizes time. Delaying action for over 12 days prevents sandbox-based security solutions—which only run tests for a few minutes—from detecting malicious behavior. This is a war of attrition, and the attackers are currently winning.
What Should You Do to Protect Yourself?
Against a campaign designed to fool both systems and humans alike, ordinary vigilance is not enough. Here are the steps you need to take starting today.
1. Only Download Apps from Official Sources
This is an absolute rule. If you want to use ChatGPT, download it from OpenAI's homepage. If you want to use Claude, download it from Anthropic's homepage. Never download from GitHub, software aggregate sites, or any other source—no matter how official it looks. Legitimate open-source projects on GitHub typically do not release compiled EXE files for end-users; if they do, they provide clear instructions on how to verify them.
2. Checking Digital Signatures Is Not Enough
As analyzed, digital signatures only confirm the origin of the executable, not the integrity of the entire package. When downloading a ZIP archive, check if there are any unfamiliar DLL files included. If a legitimate EXE file comes with an unknown DLL, that is a major red flag.
3. Beware of “Free” Versions of Paid Software
Attackers exploit the desire for high-end tools for free. If an application that normally requires payment or an official subscription is offered as a “free desktop version,” ask yourself: Why is it free? In most cases, the price isn't paid in money, but with your data and security.
4. Use Security Tools with Behavioral Detection Capabilities
Traditional signature-based antivirus solutions often miss attacks like this. Prioritize tools equipped with behavioral detection capabilities, such as tracking child processes, detecting unauthorized DLL loads, or monitoring abnormal resource consumption (a telltale sign of crypto miners).
5. Monitor for Signs of Infection
If your computer suddenly slows down abnormally, fans run constantly, or unfamiliar processes appear in Task Manager, check immediately. Crypto miners typically leave clear footprints regarding CPU and GPU usage. Do not ignore minor signs, as they could be clues pointing to a much larger attack.
When Brand Trust Becomes a Vulnerability
I have witnessed the cybersecurity industry change dramatically over the past two decades, from simple floppy-disk-borne viruses to sophisticated campaigns using legitimate digital signatures to fool systems. Every step forward taken by attackers demands a corresponding evolution in user awareness.
The Yoba campaign serves as a reminder that brand trust cannot replace verification. An EXE file signed by OpenAI or Anthropic does not mean the entire software package you downloaded is safe. And a professional-looking GitHub project does not mean it is trustworthy.
Attackers are getting increasingly adept at wearing the cloaks of reputable brands. They no longer launch frontal assaults; instead, they parasitize what you already trust. In that world, the ultimate survival skill is not trusting less, but verifying more.
Have you ever downloaded an AI app from an unofficial source just because it showed up first in search results and looked “trustworthy”? If so, what made you trust it—and knowing that attackers can borrow digital signatures from OpenAI or Anthropic to trick you, how will you change the way you download and check software? Share your thoughts or experiences below, because in the battle against these increasingly sophisticated tricks, everyone's vigilance helps make the community safer!
0 Comment
Add your comment to this article