New reports from security agencies indicate that the LiteLLM tool, one of the popular open-source libraries for AI software development, has been targeted by a supply-chain attack. This breach has led to the exposure of several terabytes of security credentials, many of which belong to the world's most sensitive organizations.
According to reports from security firms CloudSEK and Hudson Rock, the list of victims includes major names such as Microsoft, Amazon, Cisco, Samsung, and Salesforce. The leaked data includes cloud keys, repository tokens, SSH keys, Kubernetes secrets, and proprietary keys of AI providers, which could allow attackers unauthorized access to the critical infrastructure of over 2,500 organizations.
This security disaster occurred during a mere 40-minute window in March. During this time, users who downloaded infected versions of LiteLLM directly from the official PyPI repository unwittingly handed their sensitive information over to attackers. Hudson Rock finalized this discovery by analyzing a massive 195-terabyte file, highlighting the deep risks associated with AI development middleware tools.