Skip to the content
AI and Cybersecurity Learning and Practice Center Test automation Open the Learning Center

NewsIncident

LiteLLM compromised again: a poisoned 1.107.0 on PyPI and Docker stole cloud, developer and AI tool credentials

LiteLLM says its release pipeline was used on October 21 to publish version 1.107.0, built from a lookalike repository, Berri-AI/litellm, with malware that steals credentials and loads remote-code tools. The seven PyPI wheels were live from 11:39 to 13:23 UTC and the Docker tags v1.107.0, latest, main-latest and main-stable from 11:46 to 14:20 UTC. The attacker used three engineers' GitHub accounts, apparently through stolen GitHub CLI OAuth tokens: one overrode branch protection and changed the workflow, a second approved the production release and a third deleted the workflow records. The malware sits in Python startup files, so it runs whenever Python starts, and per Socket it collects AWS, Azure, Google Cloud, Kubernetes and Docker credentials, Git, SSH and GPG keys, browser logins and the login files of Codex, Claude, Gemini and Cursor, and tries LiteLLM proxy databases. Wiz saw the stolen AWS credentials tested about four and a half hours later, including checks of Bedrock model access that it links to LLMjacking. Help Net Security notes this is LiteLLM's second supply chain attack this year, after one in March. Enterprise customers on the private registry were not affected.

Why it matters

An AI gateway's package holds the keys to every model provider and cloud it touches, so one bad release is a credential breach. Pin versions, and if 1.107.0 ever ran, rebuild from 1.106.2 and rotate everything that host could reach.

Sources

Read it with the rest of the news in the Learning Center