LiteLLM 1.107.0 Ships Stealer, Backdoor
LiteLLM 1.107.0 on PyPI contained a credential stealer and remote-code-execution backdoor that ran at Python startup, harvesting cloud, CI/CD, SSH, browser, and AI-provider credentials. The compromised release was removed, but matching Docker images and floating tags also require remediation.
This is a severe AI supply-chain compromise because LiteLLM sits directly in the credential path for multi-provider model infrastructure.
- –Malicious `cook.pth` and `sitecustomize.py` executed even when applications never imported LiteLLM
- –The stealer targeted valuable secrets, including provider API keys, cloud credentials, SSH keys, and CI/CD tokens
- –An hourly loader could fetch and execute attacker-controlled Python code
- –Docker `latest`, `main-latest`, and `main-stable` tags also resolved to infected images
- –Developers should pin known-clean versions, inspect environments, rotate exposed secrets, and verify image provenance
DISCOVERED
1h ago
2026-10-21
PUBLISHED
4h ago
2026-10-21
RELEVANCE
AUTHOR
abhisek