LiteLLM 1.107.0 Release Puts Cloud Credentials and AI Keys at Risk

SlowMist warned that a malicious LiteLLM release can expose cloud credentials, AI API keys and SSH private keys, putting organizations using the AI gateway at risk of credential theft and remote code execution.

The affected version, litellm 1.107.0, was distributed through the PyPI package and Docker images, including floating tags such as latest and main-stable. The injected code runs automatically when Python starts, allowing it to collect sensitive data and support remote code execution.

SlowMist described the incident as another supply-chain attack against LiteLLM. The reported targets span credentials for cloud services, AI applications and SSH access, bringing several kinds of access keys into scope.

The alert urged users to avoid version 1.107.0, rebuild affected environments from clean images and rotate credentials that may have been exposed. It recommended pinning installations to version 1.106.2. LiteLLM is an AI gateway, and the warning covers both its Python package and Docker images.