On May 20, 2026, GitHub confirmed the compromise of internal repositories following an employee’s installation of a malicious Microsoft Visual Studio Code extension. Nearly 3,800 internal repositories were exfiltrated by the hacker group TeamPCP, which proceeded to list the source code for sale at $50,000 on a cybercriminal forum. At the time of GitHub’s initial official announcement, no impact on customer repositories had been established.
In recent years, GitHub has become a critical hub of global software infrastructure, housing access credentials for cloud environments, authentication tokens, API keys, proprietary source code, and continuous integration pipelines. For a cybercriminal group capable of breaching this platform, it presents a gateway to numerous organizations without the need to target them individually. This strategy aligns with the ongoing Mini Shai-Hulud campaign attributed to TeamPCP, which has been operational since the beginning of 2026. The group has conducted a series of cascading attacks on tools commonly used by developers, including TanStack, LiteLLM, Trivy, and the guardrails-ai package just days before affecting GitHub. Each compromise has served as a springboard for the next, with stolen access tokens and secrets facilitating further breaches. The underlying worm is designed to propagate laterally through continuous integration pipelines and package registries without requiring manual intervention.
GitHub is particularly vulnerable to such campaigns, hosting the code of over 100 million developers worldwide. Its internal repositories contain crucial source code related to the platform itself, infrastructure tools, Copilot projects, GitHub Actions workflows, and snippets from customer support interactions. While the breach of this internal layer does not affect customer repositories, it reveals sensitive information about the platform’s internal architecture, existing security mechanisms, and software dependencies that could enable more targeted future attacks. Additionally, as a subsidiary of Microsoft since 2018, GitHub operates as a U.S.-based entity under the Cloud Act.
The Entry Vector: A Poisoned VS Code Extension
On May 18, 2026, a GitHub employee’s workstation was compromised via a malicious Microsoft Visual Studio Code extension published by a third party. In its official announcement, GitHub referenced security advisory GHSA-c9j4-9m59-847w linked to the compromised Nx Console extension, a widely-used tool among developers working with the Nx framework. The Nx team acknowledged that the compromised version was published after attackers gained control of the publishing account, asserting that very few users were affected based on their assessment.
This mechanism is characteristic of software supply chain attacks targeting development environments. A VS Code extension runs with the current user’s privileges and has access to the local file system, environment variables, stored authentication tokens, and configurations of development tools. The official Microsoft marketplace does not guarantee the ongoing integrity of published extensions. A compromised developer account can transform a legitimate extension into an attack vector without any visible changes to its identifier or rating. Once the malicious extension was installed on the GitHub employee’s workstation, the attackers gained a foothold in an environment with access to the platform’s internal repositories. GitHub detected the compromise on May 18, isolated the endpoint, and promptly initiated an incident response. Critical secrets were rotated the same day, prioritizing credentials with the highest impact.
Exfiltration: 3,800 Internal Repositories in Just Hours
From the compromised workstation, TeamPCP exfiltrated approximately 3,800 of GitHub’s internal private repositories. This figure, confirmed by the group itself on the cybercriminal forum, was described as “consistent with the ongoing investigation” by GitHub in its official post. The exfiltrated repositories reportedly contained source code related to GitHub Actions, agent workflows, internal Copilot projects, CodeQL tools, internal infrastructure, security tools, marketing assets, as well as components from Codespaces and Dependabot programs.
The speed of the exfiltration echoes a recurring pattern in TeamPCP’s attacks. Once a foothold is obtained, the group acts swiftly to maximize the volume of exfiltrated data before detection. Security researchers analyzing previous attacks in the Mini Shai-Hulud campaign noted that the worm includes an automatic propagation mechanism within AWS environments via the SSM service and across Kubernetes clusters using kubectl, allowing it to move laterally without manual intervention. In GitHub’s case, the exfiltration surface was already accessible from the compromised workstation without needing further propagation. Accessing the internal repositories through the employee’s credentials was sufficient to achieve a substantial volume of exfiltration in a very short time, underscoring the importance of the principle of least privilege in internal development environments.
Extortion Attempts and GitHub’s Response
Upon listing the stolen data for sale on the cybercriminal forum, TeamPCP included an unusual caveat: the group sought to sell the source code to a single buyer for $50,000, promising to destroy their copy afterward. In the absence of a buyer, the group threatened to release the data publicly for free. Subsequently, LAPSUS$, another cybercriminal group, joined the operation, presenting a combined offer of $95,000.
GitHub steadfastly declined any negotiations, maintaining a public communication focused on verified facts. The blog post by Alexis Wales, GitHub’s Chief Information Security Officer, published on May 20, outlined the measures taken without speculating on the attackers’ intentions or the exact content of the exfiltrated data. The company announced plans to release a comprehensive report following the investigation. As of May 22, 2026, this report had not yet been published and will serve as the sole sourced reference for a detailed reconstruction of the attack chain.
IDE Extensions and Package Registries: An Underestimated Attack Surface
The attack on GitHub highlights a structural issue that security teams in organizations using IDEs like VS Code must now seriously consider. A developer installing an extension from Microsoft’s official marketplace exposes themselves to an attack surface that is challenging to control. VS Code extensions are JavaScript programs executed with user privileges, lacking strict sandboxing. They can read local files, access environment variables, interact with command-line tools, and exfiltrate data to remote servers. Integrity checks for extensions do not equate to those for signed system packages.
The Mini Shai-Hulud campaign systematically exploited this vector throughout the first half of 2026. The Python package durabletask, an official Microsoft client for the Durable Task framework, was compromised a few days after the GitHub incident by the same group. The payload embedded in versions 1.4.1, 1.4.2, and 1.4.3 of the package acted as a secret extractor capable of harvesting credentials from major cloud providers, password managers, and development tools, then exfiltrating them to an attacker-controlled server. This package is downloaded approximately 417,000 times a month. The malicious code runs automatically upon package import, without error messages or visible signs of compromise.
Three Levels of Insight
For information security officers and DevSecOps teams, the GitHub incident provides three complementary insights. The first concerns the policy for approving development extensions. Development environments must be treated as sensitive perimeters, with approved extension lists, monitoring of automatic updates, and detection of abnormal IDE process behavior. An extension that was safe yesterday may become malicious after a developer account compromise.
The second insight involves logging and detection in CI/CD pipelines and package registries. TeamPCP’s campaign successfully compromised several official packages from major vendors by exploiting stolen PyPI and npm tokens from previously compromised environments. Regular and systematic rotation of access secrets in registries, monitoring for abnormal releases, and verifying package signatures are accessible preventive measures that many organizations have yet to implement.
The third insight addresses organizations’ dependency on GitHub itself. As a subsidiary of Microsoft, which is a U.S.-based company subject to the Cloud Act, GitHub’s internal repositories may contain snippets from customer support interactions, configuration data, and infrastructure secrets. Organizations hosting sensitive business code on the platform must incorporate this dimension into their sovereign risk assessments, regardless of the ongoing incident. The promised comprehensive report from GitHub will serve as a reference point for any serious evaluation of actual exposure.



