← Back to live feed1 story
Security researchers have identified active exploitation of credentials stolen from a compromised LiteLLM Python package, a campaign tracked by Wiz as 'LLMJacking' targeting corporate infrastructure via IP addresses 168.100.8[.]57 and 165.227.201[.]8. The breach affected LiteLLM version 1.107.0 by embedding malware in all seven compiled wheels, which activates at Python startup and targets credentials for AWS, Azure, Google Cloud, and AI providers like Hugging Face and Gemini. PyPI and Docker Hub have removed the malicious releases and images, and affected users are urged to rotate all secret keys and pin their installations to version 1.106.2.
Sign in to suggest
Key sources
- SOURCE@jfrogsecurity“supply-chain attack involving malicious code in the LiteLLM 1.107.0 PyPI release”x.com
- SUPPORT@socketsecurity“Socket flagged all seven malicious wheels within seven minutes of publication”x.com
- SUPPORT@darkwebinformer“All seven LiteLLM 1.107.0 wheels were affected across Windows, macOS and Linux”x.com
- SOURCE@ramimacisabird“abuse of stolen credentials - initially related to LLMJacking - from the following IPs * 168.100.8[.]57 * 165.227.201[.]8”x.com
- SUPPORT@feross“LiteLLM, a Python package for connecting to AI model providers with 92M+ monthly downloads, has been compromised”x.com
- SUPPORT@klieret“If you're using mini-swe-agent, please make sure you didn't update to LiteLLM 1.107.0”x.com
- SUPPORT@altryne“new supply chain (meaning if your agents are installing packages that depend on packages) in LiteLLM”x.com
- SOURCEhuggingnewshuggingnews.com