LiteLLM Compromised, Steals Credentials

A malicious version (1.107.0) of the LiteLLM package was uploaded to PyPI and Docker Hub on October 21. The compromised package contains a credential stealer targeting browser-stored credentials, cloud secrets, and AI-related secrets. Additionally, the malicious version includes a dormant command-and-control (C2) capability, suggesting potential for future remote exploitation. This is a supply chain attack targeting AI/ML developers and organizations using LiteLLM for managing large language model integrations. Users who installed version 1.107.0 should consider themselves compromised and rotate all secrets immediately. The attack highlights ongoing risks in open-source AI tooling ecosystems and package registries.