Compromised LiteLLM 1.107.0 on PyPI Installs a Credential Stealer and Backdoor
Points and comments are a snapshot, not live.
LiteLLM 1.107.0 on PyPI installed credential-stealing malware and a backdoor via a compromised release pipeline.
The malicious release placed cook.pth and sitecustomize.py in site-packages, running in every Python process. The stealer collected SSH keys, cloud credentials, browser passwords, AI tool tokens, and LiteLLM proxy database secrets. It encrypted data and sent it to www.litellm[.]dev. A backdoor daemon checked an attacker's GitHub repository hourly for a server address, then downloaded and executed code. Three compromised developer credentials allowed bypassing branch protection. PyPI removed the release by 13:23 UTC. Docker images were also affected. Versions 1.106.2, 1.107.0rc1, and 1.107.0rc2 are clean.
What commenters are saying
The thread is dominated by frustration that LiteLLM was compromised for the second time, with commenters calling it malpractice. Several users debated supply-chain defenses: some advocated for default package cooldowns of 24-48 hours, others noted attackers could mark malicious releases as security updates to bypass such measures. A LiteLLM employee clarified that the separate private enterprise registry mentioned in a GitHub issue was for a different, unrelated vulnerability that only affects Enterprise deployments. Others promoted using Go-based alternatives or sandboxing tools like PMG.