The October 21, 2026, release of LiteLLM version 1.107.0 was compromised to include malicious code that automatically executes upon Python interpreter initialization, threatening developer and cloud environments.
Key Points
- Attackers compromised the LiteLLM 1.107.0 release on PyPI by hijacking developer GitHub credentials to redirect the workflow to a malicious repository.
- The malicious package uses Python startup hooks, such as
.pthandsitecustomize.pyfiles, to execute a credential collector without requiring the application to be imported. - The payload targets sensitive data including cloud credentials, SSH keys, browser history, environment variables, and AI-related configuration files.
- Collected data is encrypted using AES-256-CBC and RSA, then exfiltrated to an attacker-controlled domain via HTTP POST requests.
- Unix-based systems were also exposed to a secondary loader capable of retrieving and executing additional remote Python code.
- Microsoft advises organizations to rotate all potentially exposed credentials and rebuild environments from verified, clean artifacts rather than simply deleting the package.