Massive Credential Leak in AI Supply Chain Attack
LiteLLM compromise exposes terabytes of secrets from Microsoft, Amazon, Cisco, and thousands of other companies
August 13, 2026 · 3 min read
TL;DR: A supply chain attack on LiteLLM has compromised terabytes of credentials from over 2,500 organizations, including Microsoft and Amazon. Attackers exploited malicious versions on PyPI for 40 minutes in March. Companies should rotate credentials and review their systems.
What Happened?
On August 6, 2026, security firms CloudSEK and Hudson Rock revealed a massive supply chain attack on LiteLLM, an open-source tool widely used to simplify AI software development. Attackers compromised LiteLLM versions hosted on the official Python Package Index (PyPI) repository, and during a 40-minute window in March, users who downloaded these malicious versions had their credentials exposed.
According to Hudson Rock, analysis of a 195-terabyte file revealed an unprecedented amount of secrets: cloud keys, repository tokens, SSH keys, Kubernetes secrets, package publishing credentials, environment variables, and AI provider keys. CloudSEK estimates that over 2,500 organizations are affected, including prominent names such as Microsoft, Amazon, Cisco, Samsung, and Salesforce.
Why It Matters
This attack is one of the most severe against the AI software supply chain to date. Not only does it affect top-tier tech companies, but it also exposes the fragility of the open-source ecosystems that underpin AI development. The rapid adoption of tools like LiteLLM, which simplify access to multiple AI models, has created a huge attack surface. Attackers can use these credentials to access internal systems, steal data, implant malware, or launch subsequent attacks.
The fact that the attack went unnoticed for months underscores the lack of visibility and rapid response in many security teams. Furthermore, the scale of the leaked file (195 TB) suggests that attackers had access to a massive amount of sensitive information, which could have long-term implications.
Consequences and Lessons
The consequences of this attack are manifold. First, affected organizations must immediately rotate all compromised credentials, which can be a complex and costly process. Second, the incident erodes trust in open-source tools, which are fundamental to AI innovation. Companies will need to implement stricter security measures, such as package integrity verification and continuous access monitoring.
This attack also highlights the need for greater collaboration among package maintainers, code repositories, and the companies that use them. The open-source community must adopt more robust security practices, such as package signing and dependency auditing.
“This is a wake-up call for the AI industry: security cannot be an afterthought in software development.”
Security teams must assume that their open-source dependencies may be compromised and prepare to respond quickly. Credential rotation should be a regular practice, and monitoring for suspicious access should be continuous.
What Readers Should Know
If your organization uses LiteLLM or any similar tool, it is crucial to check whether you have been affected. CloudSEK and Hudson Rock have made tools available for companies to verify if their credentials are compromised. Additionally, it is advisable to review access logs and look for any unusual activity.
For individual developers, it is essential to be aware of the risks when downloading packages from PyPI and other repositories. Verifying package integrity and using isolated environments can reduce risk.
Conclusion
The LiteLLM attack is a reminder that supply chain security is a critical challenge in the AI era. Companies must adopt a proactive approach, investing in security tools and fostering a security culture throughout the organization. Transparency and cooperation among all ecosystem actors will be key to preventing future incidents.