Massive AI Supply-Chain Breach Exposes Terabytes of Credentials
Hey, so there’s been a major issue with LiteLLM, which is an open-source tool meant to improve AI development. It turns out that a ton of sensitive information was leaked, affecting big names like Microsoft, Amazon, Cisco, Samsung, and Salesforce. This was brought to light by security firms CloudSEK and Hudson Rock earlier this week. They found a variety of credentials, such as cloud keys and SSH keys, which could give access to more than 2,500 companies. The breach happened in March over a brief 40-minute window when users downloaded certain versions from the Python Package Index. Hudson Rock found this out while sifting through a massive 195TB file, but they still don't know where the data came from. This really highlights the serious weaknesses in AI software security.
Key facts
- Supply-chain attack on LiteLLM exposed credentials of major organizations including Microsoft, Amazon, Cisco, Samsung, and Salesforce.
- Disclosed by security firms CloudSEK and Hudson Rock on Tuesday and Wednesday.
- CloudSEK found cloud keys, repository tokens, SSH keys, Kubernetes secrets, package publishing credentials, environment variables, and AI provider keys.
- Attackers could potentially access more than 2,500 organizations.
- Credentials were extracted during a 40-minute window in March.
- Compromised versions of LiteLLM were downloaded from the Python Package Index repository.
- Hudson Rock analyzed a 195TB file to make the discovery.
- Neither firm identified the source of the information.
Entities
Institutions
- Microsoft
- Amazon
- Cisco
- Samsung
- Salesforce
- CloudSEK
- Hudson Rock
- Python Package Index