The accidental leak of the source code of Claude Code, Anthropic’s artificial intelligence-assisted programming tool, gained a new dimension of risk. In late March, a packaging error exposed more than 500,000 lines of TypeScript code from version 2.1.88. Poucos days later, malicious actors began creating fake repositories on the GitHub platform to attract developers interested in the material. Esses repositories promise unlocked or full versions of the code and actually deliver an infostealer known as Vidar.
Criminals optimize repositories to appear among the first results of searches related to the leak. Usuários who look for the original code end up downloading compressed files that contain a dropper in Rust. Essa tool installs the malware Vidar, responsible for stealing credentials, browsing history, sensitive files and other information stored on the victim’s computer. In some cases, the package also includes GhostSocks, which turns the infected device into a proxy for malicious traffic.
Fake repositories lure developers looking for leaked code
Security researchers have identified at least two repositories maintained by a user identified as “idbzoomh”. Esses spaces simulate legitimate mirror repositories of the leaked code. Eles include descriptions that mention unlocked enterprise features to increase appeal among technology professionals.
The infection process begins with the download of a ZIP file or executable disguised as an update or modified version. When executing the file, the dropper activates the Vidar silent installation mechanism. The infostealer collects data and sends it to servers controlled by the attackers, often using proxies to make tracking difficult.
More on this story: Sony disables PlayStation 4 online features to focus on PS5 expansion
The campaign directly exploits the interest generated by the initial leak, which occurred through a source map file mistakenly included in the npm package. Esse file allowed the complete reconstruction of the tool’s internal architecture, including agent and permissions components not yet publicly released.
Technical Details of Vidar and GhostSocks Malware
Vidar operates like a classic infostealer, focused on extracting valuable information without attracting too much attention from the user. Ele scans common folders, browsers and installed applications for saved passwords, cookies, API keys and documents. Integration with GhostSocks expands the reach of the threat by allowing the compromised device to serve as an intermediary in other criminal operations.
Cybersecurity experts highlight that the combination of the two malware makes the infection particularly dangerous for developers. Muitos professionals store access credentials to repositories, cloud environments and corporate accounts on their work computers. The theft of this data can lead to chain compromises of projects and infrastructures.
Distribution occurs mainly through organic searches on Google and direct navigation on GitHub. Fake repositories gain quick visibility because they replicate names and structures similar to the legitimate mirrors that emerged shortly after the leak.
Anthropic confirms cause of initial leak
Anthropic publicly acknowledged the incident as a packaging issue caused by human error. The company clarified that the approximately 60 MB font map file was included by accident in the version published in the npm registry. Nenhum Customer data, credentials or weights of AI models were exposed in the episode.
The leak revealed details of Claude Code’s CLI implementation, including multi-tasking agent logic and features in development. Embora the source code does not include sensitive user information, it offers visibility into the internal architecture of the tool, which attracts both researchers and malicious actors.
The company acted quickly to remove the problematic package and mitigate exposure. Mesmo Thus, copies of the code had already been downloaded and spread across several public repositories before the fix.
Learn more: Accidental leak exposes full source code of Anthropic’s Claude Code
Risks for the developer community
Developers seeking to study leaked code face the risk of installing modified versions without adequate verification. The practice of downloading files from unofficial sources significantly increases the chance of infection, especially in a scenario of high technical curiosity generated by the incident.
Using up-to-date antivirus software and checking file hashes before execution represent basic protection measures. Especialistas recommend avoiding repositories that promise “unlocked” or “enterprise” versions of the code, as these offers often hide malicious intent.
The speed at which the fake repositories gained traction demonstrates how the collective interest in AI tool leaks can be exploited within hours. Plataformas like GitHub rely on community vigilance to identify and remove malicious content, but the initial responsibility lies with the end user.
Recommended preventative measures in the npm and GitHub ecosystem
Technology professionals should prioritize installing packages directly from official sources whenever possible. In the case of leaked code, best practice involves waiting for independent reviews before any download. Verificar the reputation of the repository maintainer and analyzing the contents of files with security tools also helps reduce risk.
The episode reinforces the importance of rigorous packaging processes in companies that deal with sensitive technologies. Erros Humans in release pipelines continue to be one of the main entry points for unintentional exposures of internal code.
The cybersecurity community is closely following the evolution of this campaign. Relatórios indicates that Vidar continues to be active in other independent operations, which suggests that attackers may expand the use of the Claude Code leak as an additional distribution vector.
Full coverage: Latest News (EN)
- Avoid clicking on repository links that appear in leak-related search results without prior verification.
- Use official package managers and keep security tools always up to date.
- Perform regular system scans after any download of source code of dubious origin.
- Prefer isolated environments or virtual machines for analyzing code obtained from public sources.
The case illustrates the security challenges in the artificial intelligence development ecosystem. Enquanto The initial leak exposed only CLI tool code, subsequent exploitation transformed the legitimate interest into an opportunity to steal real user data.
Anthropic continues working to strengthen its internal release processes. Desenvolvedores and companies that use similar tools should double their attention to digital hygiene practices, especially during periods of high movement of technical content on the internet.