On March 24, 2026, two versions of the AI gateway package litellm on PyPI were backdoored. Versions 1.82.7 and 1.82.8 stole SSH keys, cloud credentials and Kubernetes secrets from the machines that installed them. Anyone who installed litellm that day without pinning a safe version may be affected.
What is going on
PyPI is the main library shop for Python software. Attackers, dubbed TeamPCP, obtained the litellm maintainer's PyPI credentials through an earlier compromise of the Trivy CI system. With those credentials they published two poisoned releases under the real project name.
Version 1.82.8 went a step further. It added a file called litellm_init.pth, which runs automatically every time Python starts on that machine. Think of it as a spare key left in a lock: the burglar can walk back in long after the original delivery was thrown away.
This is a supply chain attack. Nothing was wrong with your own code; the trusted ingredient was swapped out upstream.
Who is affected
Affected are systems that ran pip install litellm on March 24, 2026 without pinning to a known-good version. That includes build servers, containers and developer laptops.
For small businesses, the realistic path is an AI feature on your website: a chatbot, a product description generator or a support assistant built on litellm. If a contractor or agency set that up for you, ask them directly. Versions 1.82.6 and earlier, and 1.83.0 and later, are not the poisoned builds.
What to do now
- Pin litellm to 1.82.6 or to 1.83.0 or later, then reinstall on every affected machine and container.
- Search for and delete any
litellm_init.pthfile. It runs on every Python start, so removing the package alone is not enough. - Rotate all credentials the machine could reach: LLM API keys, cloud keys, database passwords and Kubernetes secrets.
- Replace SSH keys that were stored on affected systems and remove the old public keys from anything that trusted them.
- Rebuild containers and CI images from scratch rather than patching them in place.
How to check your own site
This incident started in a package registry, not in your web server, so no website scan can confirm it on its own. What a scan can do is show what an attacker with stolen keys would find next. The free Website Security Check flags outdated CMS versions, known-vulnerable plugins, missing security headers, weak TLS and email-authentication gaps, and the Full scan adds exposed-file and secret checks on sites whose ownership you have verified.
Source: LiteLLM security update, March 2026.
Comments
No comments yet.