18 Nov 2026, Wed

Massive "wfx-loot" Campaign Exploits Stolen Credentials to Target Over 60,000 GitHub Repositories

In a sophisticated and widespread security operation that has sent shockwaves through the software development community, an unidentified attacker has leveraged stolen access to compromise tens of thousands of GitHub repositories. According to reports from multiple leading cybersecurity firms, the campaign—dubbed "wfx-loot"—began on November 10 and has targeted over 60,000 repositories, systematically testing for vulnerabilities and injecting malicious code designed to harvest sensitive credentials.

The scale of the operation is significant, highlighting the ongoing risks associated with compromised developer accounts and the potential for automated abuse of Continuous Integration (CI) systems. Aikido Security, which has been tracking the campaign, reported that it identified roughly 4,800 individual accounts that had been compromised. These accounts were utilized to submit "probe" pull requests to 61,764 distinct repositories. This massive, automated outreach served as a reconnaissance phase, allowing the attacker to determine which repositories were susceptible to unauthorized code injection based on existing permission configurations.

Understanding the Mechanism of the Attack

Industry experts, including those at StepSecurity, have emphasized that the pull requests themselves were not inherently malicious. Instead, they functioned as a litmus test for the attacker to confirm whether their stolen access credentials permitted them to modify the repository’s code or workflow configurations. Once the attacker confirmed that a repository was vulnerable, they moved to the next phase: the deployment of malicious payloads.

The tactical approach taken by the wfx-loot threat actor is notable for its avoidance of traditional software exploits. As StepSecurity pointed out in its analysis, the intruder is not exploiting a bug or vulnerability within GitHub Actions itself. Rather, the campaign relies entirely on the abuse of legitimate, stolen tokens belonging to developers or automated bot accounts. By operating within the parameters of authorized access, the attacker is able to evade many standard security triggers that are designed to look for external intrusions.

The breadth of the impact remains a subject of ongoing investigation. While Aikido Security identified clear evidence that the credential-stealing malware successfully executed in more than 1,800 repositories, other firms have noted even higher numbers of potential exposure. SafeDep, for instance, counted 8,093 public repositories that received malicious commits. It is important to note that the receipt of a malicious commit does not necessarily guarantee that the code was successfully executed or that the payload was fully deployed; however, the volume of affected repositories underscores the audacity and reach of the attacker.

Stealth and Persistence: The Anatomy of the Stealer

The malware deployed in this campaign is designed for high-impact data exfiltration and long-term persistence. The attacker typically placed the malicious code within existing or newly created GitHub Actions workflows, shell scripts that are automatically executed by CI systems, or npm installation hooks. The use of npm hooks is particularly concerning, as it introduces a secondary risk to developers who may inadvertently install compromised dependencies while the malicious changes remain present in the repository.

Once executed, the stealer is programmed to perform a comprehensive search for high-value data. It scans the environment for GitHub tokens, repository secrets, cloud infrastructure credentials, and keys associated with various AI services. Beyond simple file scanning, the malware has the capability to read the memory of GitHub Actions runner processes, allowing it to extract sensitive information that might otherwise be protected. Furthermore, the script includes functionality designed to expand the attacker’s footprint by establishing persistent access within AWS and Azure cloud environments.

To avoid detection by repository owners and automated monitoring tools, the attacker often engaged in a practice of "cleaning up" after themselves. In many instances, the intruder would reverse the malicious code changes and delete the workflow runs once the desired data had been exfiltrated, effectively masking the evidence of their activity.

The Mystery of the Originating Tokens

One of the more perplexing aspects of the wfx-loot campaign is the source of the initial access. StepSecurity successfully traced 173 of the probe pull requests to tokens that were originally issued to the Fly.io GitHub app. This discovery prompted immediate concern regarding the security of the Fly.io infrastructure. However, researchers have been unable to determine how these tokens were obtained in the first place, and there is currently no evidence to suggest that the Fly.io platform itself suffered a direct breach. This suggests that the tokens may have been harvested through other means, such as secondary compromises or third-party leaks, before being repurposed for this large-scale operation.

The impact of the campaign has already manifested in the real-world software supply chain. Affected software releases have been identified, including a compromised Packagist package and specific versions of two Go modules. This realization serves as a stark reminder of how quickly a compromised developer account can cascade into a widespread supply chain vulnerability, affecting downstream users who rely on these packages.

Comparing Threats and Recommendations for Mitigation

Researchers at Aikido Security have noted distinct similarities between the wfx-loot operation and the "GhostAction" campaign that emerged in October. Despite these technical parallels, Aikido has assessed that the two operations are likely the work of different threat actors, suggesting that this particular method of abusing GitHub Actions has become a popular blueprint for malicious activity.

The cybersecurity community is urging organizations to treat this situation with the utmost seriousness. Because of the nature of the malware—which specifically targets secrets and environment variables—security firms have warned that any organization where the malware was successfully executed should assume that all accessible secrets have been compromised.

The recommended course of action for affected organizations is comprehensive and immediate. Security teams are advised to revoke all potentially compromised tokens, rotate all exposed secrets, and conduct a thorough audit of cloud infrastructure to identify any unauthorized changes or persistent access points established during the window of exposure.

As the investigation into wfx-loot continues, the incident serves as a significant wake-up call for the development industry. It reinforces the necessity of strict access controls, the importance of limiting the permissions granted to CI/CD workflows, and the critical need for constant monitoring of repository activity. The ability of a single attacker to systematically probe and compromise over 60,000 repositories using nothing more than stolen, legitimate tokens highlights a persistent vulnerability in the modern software development lifecycle—a vulnerability that will likely continue to be exploited until organizations adopt more rigorous identity management and secret protection protocols.

The incident follows a string of high-profile security challenges facing the development ecosystem. From the compromise of customer data at Fireworks AI to the ongoing concerns regarding active credentials left exposed on public repositories, the wfx-loot campaign is part of a larger, systemic struggle to secure the tools and environments that power the modern digital economy. With the threat landscape evolving, the focus for developers and security professionals must shift toward proactive threat hunting and the assumption of breach, ensuring that when credentials are stolen, the blast radius of any subsequent attack is as contained as possible.

Leave a Reply

Your email address will not be published. Required fields are marked *