LiteLLM's compromised release workflow sent credential-stealing malware to PyPI and Docker
The October 21st attack used a lookalike source repository and credentials from three developer accounts to publish version 1.107.0. LiteLLM suspects OAuth tokens leaked through logs in an internet-exposed deployment, but has not confirmed how that deployment was accessed.
By RuntimeWire Staff · Published
Primary source: BleepingComputer
Why it matters
An AI gateway can sit beside model-provider keys, cloud credentials and deployment secrets. This attack showed how a compromised release workflow can carry malware through trusted software distribution and into environments holding those assets, even when the application's own code is untouched.
LiteLLM's release pipeline published malware to PyPI and container registries on October 21st, putting credentials and secrets at risk for users who installed or ran version 1.107.0. Co-founders Krrish Dholakia and Ishaan Jaffer built the open-source AI gateway to give developers one interface for calling multiple model providers. The compromised release turned that trusted point of access into a potential route to customers' cloud accounts, AI keys and developer environments.…
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.