November 2, 2026 | Digital Assets Observer
Cybersecurity researchers uncover a malware campaign exploiting trusted AI brands, GitHub search visibility and legitimate signed software components to compromise Windows devices.
The rapid adoption of artificial intelligence applications is creating new opportunities for cybercriminals to exploit users searching for productivity tools, desktop assistants and advanced AI software.
Check Point Research has identified a malware campaign known as Yoba that uses fraudulent GitHub projects and websites impersonating ChatGPT and Claude desktop applications.
The campaign is particularly concerning because attackers reportedly combined malicious files with legitimate, digitally signed executables associated with OpenAI and Anthropic.
According to the researchers, the technique allows malicious software to exploit the apparent credibility of trusted applications without compromising the AI companies themselves.
The findings highlight an emerging cybersecurity challenge: the growing difficulty of distinguishing authentic AI software downloads from convincing malicious imitations.
How the Yoba Malware Campaign Begins
The attack starts with online searches.
Cybercriminals reportedly created GitHub repositories and GitHub Pages websites advertising supposed desktop versions of advanced AI applications.
These projects used names associated with popular AI systems and directed visitors toward downloadable ZIP archives.
The malicious repositories appeared prominently in Google Search results, according to Check Point Research.
Researchers assessed the campaign as consistent with search engine optimization poisoning, commonly known as SEO poisoning.
This technique involves placing malicious websites or files where users are likely to encounter them while searching for legitimate software.
Unlike traditional phishing attacks that rely on unsolicited emails or messages, SEO poisoning exploits a user’s existing intention to find a particular product.
The victim may therefore approach the malicious page voluntarily.
Why GitHub Can Make Fake Software Appear Credible
GitHub is widely used by software developers, technology companies and open-source communities.
Its reputation as a legitimate development platform can make projects hosted there appear trustworthy to less experienced users.
However, the presence of software on GitHub does not establish that it is authentic, secure or endorsed by the company whose name appears in the repository.
Attackers can create repositories, documentation pages and release archives that resemble genuine software distribution channels.
In the Yoba campaign, fraudulent projects reportedly used AI-related branding and download links to encourage users to install malicious packages.
This demonstrates how legitimate infrastructure can be misused as part of a social engineering attack.
Understanding DLL Sideloading
One of the most technically significant findings concerns a method known as dynamic-link library, or DLL, sideloading.
Windows applications frequently load DLL files to access supporting functions.
Under certain conditions, an application may load a DLL from its local directory.
Attackers can exploit this behaviour by placing a malicious DLL alongside an otherwise legitimate executable.
When the executable runs, it may load the malicious library.
Check Point Research reports that Yoba packages used legitimate signed executables together with a malicious file named winmm.dll.
This approach creates an important distinction.
The legitimate executable may carry a valid digital signature, but the overall package can still be malicious.
Users should therefore avoid assuming that every file distributed alongside a signed application is trustworthy.
The Role of YobaDropper and YobaLoader
The malware campaign reportedly involves multiple stages.
The first stage, YobaDropper, performs checks before deploying additional components.
Researchers observed behaviours intended to complicate automated malware analysis, including execution checks, delays and regional language checks.
The dropper also seeks administrator privileges.
Once its conditions are satisfied, it deploys YobaLoader and establishes persistence through a scheduled Windows task.
Persistence allows malicious software to remain active or restart after a system reboot.
YobaLoader then provides a mechanism for retrieving and executing additional payloads.
This multi-stage architecture is important because the initial malicious file may not contain every component ultimately used during the attack.
Telegram, DNS-over-HTTPS and External Hosting
The campaign also demonstrates how attackers can misuse ordinary internet services.
According to Check Point Research, YobaLoader retrieves information associated with a Telegram channel’s profile image.
That information contributes to generating a domain used in the malware’s communication process.
The loader then uses Google’s DNS-over-HTTPS service to request DNS information associated with the generated domain.
After additional validation and processing, the malware can retrieve an encrypted payload from external hosting infrastructure.
Researchers identified Catbox among the services used in the delivery chain.
The significance is not that these services are inherently malicious.
Rather, attackers can combine legitimate online services to create a communication mechanism that is harder to identify through simple domain-blocking strategies.
Cryptocurrency Mining Payload Identified
Check Point Research observed XMRig version 6.26.0 as the final payload in the campaign it analyzed.
XMRig is cryptocurrency mining software that can be used legitimately when installed and operated with authorization.
In this case, researchers reported its unauthorized deployment through malware.
Such activity is commonly described as cryptojacking.
Unauthorized mining can consume computing resources, increase electricity usage and reduce device performance.
Researchers also observed behaviour designed to reduce visible disruption while a user was actively operating the affected computer.
However, the risk extends beyond cryptocurrency mining.
YobaLoader is capable of delivering other executable or PowerShell payloads.
Consequently, a device affected by Yoba should be treated as potentially compromised beyond the specific mining software identified in this investigation.
Were ChatGPT or Claude Hacked?
The research does not establish that ChatGPT, Claude, OpenAI or Anthropic were compromised.
This distinction is essential.
The attackers reportedly abused legitimate signed application components and used familiar AI product branding to make malicious downloads appear credible.
The attack targets software distribution trust rather than demonstrating a breach of the underlying AI platforms.
Users should therefore avoid interpreting the findings as evidence that legitimate ChatGPT or Claude applications are inherently infected.
What Businesses Should Do to Reduce the Risk
Organizations adopting AI applications should establish clear procedures for software installation and verification.
Employees should obtain applications from official vendor-controlled websites or approved enterprise distribution channels.
IT administrators should restrict unnecessary administrator privileges and maintain application allowlists where practical.
Endpoint security systems should monitor suspicious DLL loading, unauthorized scheduled tasks and unusual network activity.
Organizations should also educate employees about misleading search results and fraudulent repositories.
For businesses already concerned about a possible infection, a qualified security team should isolate affected endpoints, investigate persistence mechanisms and assess whether additional malicious payloads were delivered.
DAO Analysis: AI Adoption Is Expanding the Software Supply-Chain Attack Surface
The Yoba campaign highlights an important consequence of rapid AI adoption.
As employees independently experiment with new tools, organizations may lose visibility into the software being downloaded and installed.
Cybercriminals can exploit this behaviour without attacking AI models directly.
The use of signed executables, fraudulent repositories and trusted online services also illustrates why traditional reputation-based security checks may be insufficient.
Organizations need to verify software provenance and monitor application behaviour throughout the execution chain.
The central lesson is that safe AI adoption requires more than protecting prompts and confidential business information.
It also requires controlling how AI applications enter corporate devices.
Frequently Asked Questions
What is Yoba malware? Yoba is a multi-stage malware campaign identified by Check Point Research that uses fraudulent AI software downloads.
Does Yoba infect legitimate ChatGPT or Claude services? The research does not indicate that the official AI services were compromised.
What is DLL sideloading? It is a technique in which an application loads a malicious DLL placed in a location where the application expects to find a supporting library.
What payload was identified? Researchers observed unauthorized deployment of XMRig cryptocurrency mining software.
How can users stay safe? Download AI software from official sources, avoid suspicious repositories and keep endpoint protection updated.
Editorial note: The technical findings are attributed to Check Point Research’s October 29, 2026 communication. The article does not claim independent malware testing by DAO.
Leave a Reply