LiteLLM supply chain compromise: What you need to know
Attackers compromised LiteLLM (23 million weekly downloads) on PyPI and Docker by using stolen GitHub OAuth tokens from three engineers to launch the release pipeline against a malicious fork named Berri-AI/litellm (vs. legitimate BerriAI/litellm). The malicious version 1.107.0 injected cook.pth, cook.py, and sitecustomize.py into site-packages, executing on every Python interpreter startup regardless of whether LiteLLM was imported. The payload steals cloud credentials (AWS, GCP, Azure, Kubernetes), developer secrets (SSH, GPG, git/npm/PyPI tokens), and LLM API keys, exfiltrating them via AES-256 and RSA-4096 encrypted channels to a lookalike domain. A second-stage RAT capability fetches and executes arbitrary Python code from a GitHub repository, marking the second LiteLLM supply chain compromise in 2026.