Attackers Compromise LiteLLM Release Pipeline to Publish Credential-Stealing Package

· Original article ↗

Summary

Attackers compromised LiteLLM’s release pipeline and published version 1.107.0 with credential-stealing and remote-code execution capabilities. Wiz reported subsequent abuse of stolen AWS credentials; the initial access route remains unclear.

Key points

  • On October 21, 2026, attackers used compromised release credentials to publish LiteLLM 1.107.0 to PyPI and Docker Hub.
  • The package targeted users who installed an unpinned version during the stated window; Docker users of specified 1.107.0 and latest-related tags were also potentially affected.
  • Malicious Python files collected system, browser, developer, cloud, and AI tool secrets and sent data to litellm[.]dev/healthcheck.
  • A backdoor could periodically fetch and execute a Python payload using a GitHub-hosted domain file; the relevant GitHub content was removed.
  • Wiz reported that attackers later validated stolen AWS credentials and performed activity consistent with LLMjacking, along with some IAM enumeration and discovery.
  • LiteLLM suspects publicly exposed deployment logs may have exposed GitHub CLI OAuth tokens, but the initial access route has not been established.
  • LiteLLM says it hardened its release process. Developers are advised to remove version 1.107.0, isolate affected hosts, investigate for malicious artifacts, rotate credentials, and rebuild.

Article Details

Victim Organization
LiteLLM
Incident Type
Compromise of LiteLLM's release pipeline to publish a malicious Python package and Docker image containing a credential stealer and backdoor.
Incident Date
2026-10-21
Data Types Exposed
  • System secrets
  • Browser secrets
  • Developer credentials
  • Cloud credentials
  • AI tool secrets
  • GitHub CLI OAuth tokens
Operational Impact
LiteLLM 1.107.0 was published to PyPI and Docker Hub. Users who installed the unpinned package or used specified Docker tags during the affected windows were impacted. The malware was designed to collect credentials and keys, exfiltrate them, and execute remote code. Wiz reported subsequent validation and use of stolen AWS credentials.
Claim Status
confirmed

People

Products

Related Articles