A poisoned LiteLLM, version 1.107.0, hit PyPI and Docker on Oct 21 and steals cloud keys and AI keys.

WHAT HAPPENED
Attackers took over LiteLLM's release pipeline and added a credential stealer at build time. The official GitHub tag has no malicious files.

The bad build was on PyPI for about 2 hours (11:39 to 13:23 UTC) and on the Docker latest tags for about 2.5 hours (11:46 to 14:20 UTC).

Once installed, it runs every time Python starts in that environment, even if your code never imports LiteLLM. It goes after cloud, developer and AI provider credentials, including the keys stored in the LiteLLM proxy.

IF YOU PULLED 1.107.0
1. Treat the host as compromised.
2. Rebuild it from a clean image.
3. Rotate every secret it could reach.
4. Pin to version 1.106.2.

No CVE has been assigned and there is no confirmed victim count. The investigation is still ongoing.

SOURCES
LiteLLM advisory: https://docs.litellm.ai/blog/security-incident-oct-21-2026
Datadog Security Labs: https://securitylabs.datadoghq.com/articles/litellm-compromised-pypi-docker/
Endor Labs: https://www.endorlabs.com/learn/litellm-1-107-0-compromised-python-startup-hooks-steal-cloud-ai-and-proxy-credentials