- Joined
- Jan 20, 2026
- Messages
- 345
- Reaction score
- 2,258
When hackers are tired of holding servers, they moved to GitHub. And no one noticed.
When GitHub published another source code, it unexpectedly attracted the attention of security experts. In recent days, repositories with the name of Miasma-Open-Source-Release have begun to appear on the platform. According to the authors of исследованияthe study, most of these repositories posted through previously compromised developer accounts. One of them was soon removed, but experts managed to preserve its contents and study in detail.
A study of the code showed that Miasma is not just a self-spreading malicious code for attacks on software supply chains. The authors of the project have created a full-fledged set of tools that can use stolen accounts for attacks on packages in PyPI, NPm and RubyGems, GitHub repositories, GitHub’s assembly systems, JFrog Artifactory’s corporate storage, artificial intelligence development environments and cloud services.
According to the documentation within the project, Miasma is written on TypeScript and is designed to run both on developer computers and in continuous integration and delivery systems. The code collects secrets, spreads through software packages and repositories, and also uses SSH access and the AWS Systems Manager service to then navigate through the infrastructure.
One of the most unusual features was almost complete dependence on GitHub as a command center. Instead of their own servers, attackers use search over open-comments to transfer commands, search for stolen access keys and get updates. This approach allows you to hide malicious activity among the usual traffic to the popular development platform.
The researchers found three independent channels of control. Everyone uses their own search bar, separate cryptographic keys and different command delivery mechanisms. One channel is designed to search for stolen GitHub tokens, the other allows you to perform arbitrary JavaScript code, and the third regularly downloads and launches scripts in Python.
Miasma collects accounts from a large number of sources. Among them are Amazon Web Services, Microsoft Azure, Google Cloud, Kubernetes, HashiCorp Vault, as well as password managers 1Password and Bitwarden. If there are appropriate rights, the malicious code may look for secrets in GitHub Actions environments and even extract data directly from the memory of the working nodes.
The authors reviewed in detail how malicious code is distributed through software packages. If he finds a valid publication token, then downloads a legitimate package, implements his own load, increases the version number and publishes the infected build back in the registry. For npm, support for digital confirmations of the origin of the Sigstore package is provided, which helps to mask the swapped versions of the legitimate ones.
In addition, Miasma is able to infect more than a dozen popular programming tools with support for artificial intelligence, including Claude, Gemini, Cursor, Copilot, Kiro and Cline. To do this, the malicious code changes the configuration files, forcing assistants to execute additional commands when starting working sessions.
The authors of the project also provided for the mechanisms that allow you to gain a foothold in the system. A permanent background module can be installed on the infected machine, which every hour turns to GitHub in search of new commands. At the same time, the program tries to avoid launching on devices where popular endpoint protection tools are found.
The most aggressive function was the so-called “emergency switch”. If attackers use the victim’s stolen GitHub token to transmit data, malicious code may establish a separate monitoring of its status. If the token is withdrawn, the command that removes the user's home director and the contents of the folder with the documents will start.
The developers of the study note that Miasma demonstrates a noticeable evolution of attacks on software supply chains. Instead of a single malicious package, the attackers have created a multifunctional platform that can simultaneously steal credentials, disseminate through trusted development channels, bypass protective mechanisms and use popular cloud services to further develop the attack.
When GitHub published another source code, it unexpectedly attracted the attention of security experts. In recent days, repositories with the name of Miasma-Open-Source-Release have begun to appear on the platform. According to the authors of исследованияthe study, most of these repositories posted through previously compromised developer accounts. One of them was soon removed, but experts managed to preserve its contents and study in detail.
A study of the code showed that Miasma is not just a self-spreading malicious code for attacks on software supply chains. The authors of the project have created a full-fledged set of tools that can use stolen accounts for attacks on packages in PyPI, NPm and RubyGems, GitHub repositories, GitHub’s assembly systems, JFrog Artifactory’s corporate storage, artificial intelligence development environments and cloud services.
According to the documentation within the project, Miasma is written on TypeScript and is designed to run both on developer computers and in continuous integration and delivery systems. The code collects secrets, spreads through software packages and repositories, and also uses SSH access and the AWS Systems Manager service to then navigate through the infrastructure.
One of the most unusual features was almost complete dependence on GitHub as a command center. Instead of their own servers, attackers use search over open-comments to transfer commands, search for stolen access keys and get updates. This approach allows you to hide malicious activity among the usual traffic to the popular development platform.
The researchers found three independent channels of control. Everyone uses their own search bar, separate cryptographic keys and different command delivery mechanisms. One channel is designed to search for stolen GitHub tokens, the other allows you to perform arbitrary JavaScript code, and the third regularly downloads and launches scripts in Python.
Miasma collects accounts from a large number of sources. Among them are Amazon Web Services, Microsoft Azure, Google Cloud, Kubernetes, HashiCorp Vault, as well as password managers 1Password and Bitwarden. If there are appropriate rights, the malicious code may look for secrets in GitHub Actions environments and even extract data directly from the memory of the working nodes.
The authors reviewed in detail how malicious code is distributed through software packages. If he finds a valid publication token, then downloads a legitimate package, implements his own load, increases the version number and publishes the infected build back in the registry. For npm, support for digital confirmations of the origin of the Sigstore package is provided, which helps to mask the swapped versions of the legitimate ones.
In addition, Miasma is able to infect more than a dozen popular programming tools with support for artificial intelligence, including Claude, Gemini, Cursor, Copilot, Kiro and Cline. To do this, the malicious code changes the configuration files, forcing assistants to execute additional commands when starting working sessions.
The authors of the project also provided for the mechanisms that allow you to gain a foothold in the system. A permanent background module can be installed on the infected machine, which every hour turns to GitHub in search of new commands. At the same time, the program tries to avoid launching on devices where popular endpoint protection tools are found.
The most aggressive function was the so-called “emergency switch”. If attackers use the victim’s stolen GitHub token to transmit data, malicious code may establish a separate monitoring of its status. If the token is withdrawn, the command that removes the user's home director and the contents of the folder with the documents will start.
The developers of the study note that Miasma demonstrates a noticeable evolution of attacks on software supply chains. Instead of a single malicious package, the attackers have created a multifunctional platform that can simultaneously steal credentials, disseminate through trusted development channels, bypass protective mechanisms and use popular cloud services to further develop the attack.
