LiteLLM supply-chain breach exposes 434,000 pipelines and company secrets

A massive supply-chain attack on LiteLLM, a popular open-source tool for AI development, has exposed terabytes of credentials belonging to some of the world's largest organizations. Security researchers at CloudSEK and Hudson Rock reported the breach on Tuesday and Wednesday, noting that compromised versions of LiteLLM downloaded from the Python Package Index (PyPI) between March contained malicious code that scraped memory and exfiltrated data via an attacker-controlled channel. The 40-minute window was enough to extract credentials from over 2,500 organizations, leading to the exposure of more than 434,000 CI/CD (continuous integration/continuous delivery) pipelines. Affected entities include Nvidia, AWS, Samsung, Salesforce, Cisco, Airbus, John Deere, and the London Stock Exchange Group, among many others.

Hudson Rock obtained and analyzed a 195TB file filled with the stolen data, which contained cloud keys, repository tokens, SSH keys, Kubernetes secrets, package publishing credentials, environment variables, and AI provider keys. The research teams also found secrets for platforms like Salesforce, Slack, and Microsoft Azure. The attack, attributed to a teenage hacking group known as TeamPCP, expanded beyond LiteLLM — the vulnerability scanner Trivy, KICS, and the Telnyx Python SDK were also compromised in the same campaign. Both security firms warned that many organizations remain unaware their credentials are exposed, as the dumped variables lack identifiable company details. Independent researcher Kevin Beaumont confirmed the data's legitimacy, calling it a "massive supply chain breach" caused by poor AI security and DevOps practices.