Source: stepsecurity.io
litellm version 1.107.0 distributed via PyPI and Docker was found to contain malicious code in the form of a .pth file that acts as a credential stealer and an hourly remote code loader. The malware exfiltrates stolen credentials to litellm.dev, a domain controlled by the attacker. Both PyPI wheels and Docker images were affected, making this a significant supply chain compromise. The backdoor loads remote code on an hourly basis, enabling persistent and evolving malicious activity. StepSecurity published an analysis including indicators of compromise (IOCs) and remediation guidance. This incident highlights the ongoing risk of supply chain attacks targeting popular open-source AI/ML tooling packages. Users of litellm 1.107.0 are urged to update immediately and audit their environments for signs of compromise.
