LiteLLM suffered a supply chain attack on October 21, resulting in over 49,000 downloads of a malicious version containing information-stealing malware and remote access trojan capabilities.
Key Points
- Attackers compromised LiteLLM version 1.107.0 by hijacking GitHub CLI OAuth tokens to publish poisoned packages to PyPI and Docker Hub.
- The malicious code remained active for approximately three hours before being removed, with enterprise customers using private repositories remaining unaffected.
- The malware harvests sensitive data, including cloud secrets, SSH keys, browser credentials, and API keys for AI services like OpenAI and Anthropic.
- Cybersecurity firm Wiz reported that attackers used stolen AWS credentials to perform LLMjacking by querying Amazon Bedrock models.
- Users who installed the compromised version are advised to immediately rotate all credentials and scan their systems for persistent malware.