LiteLLM Supply-Chain Attack Plants Credential-Stealing Malware
Original source
LiteLLM release pipeline hacked to distribute info-stealer malware
BleepingComputer →Attackers compromised LiteLLM’s release pipeline on October 21, injecting malware into version 1.107.0 across PyPI wheels and Docker images published to Docker Hub and GitHub Container Registry. The Python SDK and AI gateway, with 61,800 GitHub stars and 92 million monthly downloads, serves enterprises including NVIDIA, Netflix, AT&T, and Okta. Security researchers detected the attack within six minutes, but the malicious packages remained accessible for over an hour on multiple registries.
The attack worked by placing three files (cook.pth, cook.py, and sitecustomize.py) in Python’s library directory. During startup, Python processes .pth executable lines and imports sitecustomize, triggering malware execution before any application launches. One component establishes persistence by checking a GitHub-hosted pointer every hour to download and execute code remotely. The payload harvests credentials across multiple attack surfaces: cloud platforms (AWS, Azure, Google Cloud, Kubernetes, Docker), AI provider authentication, developer secrets (Git, GitHub, npm, PyPI credentials, SSH and GPG keys), browser data, and system information. Stolen data is encrypted using AES-256-CBC with RSA-4096 before upload to a command-and-control server.
To recover, affected organizations must isolate systems and rebuild from clean backups, as simply deleting the LiteLLM package may leave malicious startup files in place. The maintainer recommends downgrading to version 1.106.2. This is the project’s second supply-chain incident; in March, the TeamPCP group published malicious versions after compromising Aqua Security’s Trivy vulnerability scanner in the CI/CD pipeline.
Read the full article
Continue reading at BleepingComputer →This is an AI-generated summary. Read the original for the full story.