How a Poisoned Security Scanner Became the Key to Backdooring LiteLLM

· Source: Blog RSS Feed | Snyk · Field: Technology & Digital — Cybersecurity & Data Privacy, Software Development & Engineering, Artificial Intelligence & Machine Learning · Depth: Advanced, long

Summary

On March 24, 2026, malicious versions 1.82.7 and 1.82.8 of the `litellm` Python package were found on PyPI, published by TeamPCP. The threat actor obtained LiteLLM's PyPI credentials by compromising a Trivy GitHub Action within LiteLLM's CI/CD pipeline. These versions, available for approximately three hours, contained a three-stage payload: a credential harvester, encrypted exfiltration to `models.litellm.cloud`, a persistent `sysmon.py` backdoor, and a Kubernetes worm. Version 1.82.7 injected code into `proxy_server.py`, while 1.82.8 used a `.pth` file for execution on every Python interpreter startup, leading to a fork bomb that aided discovery. This incident is Phase 09 of an ongoing TeamPCP campaign, which previously targeted Trivy and Checkmarx KICS, utilizing consistent attack infrastructure.

Key takeaway

For MLOps Engineers or AI Security Engineers managing Python dependencies and CI/CD pipelines, this incident underscores the critical risk of unpinned versions. If your systems installed `litellm` 1.82.7 or 1.82.8, immediately follow the remediation steps: check for persistence artifacts, rotate all affected credentials (SSH, cloud, API keys), and install a clean version on a fresh environment. Proactively, you must pin all CI/CD dependencies to exact versions to prevent similar supply chain compromises.

Key insights

Compromised CI/CD tools can enable supply chain attacks by exfiltrating publisher credentials to inject malicious code into legitimate packages.

Principles

Method

Attackers exploited a poisoned CI/CD scanner to steal PyPI credentials, then published malicious `litellm` versions with a multi-stage payload for credential harvesting, encrypted exfiltration, and Kubernetes lateral movement.

In practice

Topics

Code references

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Security Engineer, MLOps Engineer, Software Engineer

Related on AIssential

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Blog RSS Feed | Snyk.