Skip to content
NewA calmer, faster ZeroHour

Malicious LiteLLM 1.107.0 published with a credential stealer

Original title: LiteLLM supply chain compromise: What you need to know

Datadog Security Labsby @datadoghq ()3 min read
Part of a story covered by 16 sourcesTrojanized LiteLLM 1.107.0 removed; users urged to rotate secretsMerged summary & timeline
AI summary · x-ai/grok-4.7Summaries by Vani AI

PyPI and official Docker tags carried the release for about two hours; the RAT dead-drop never delivered a live second stage.

On October 21, 2026, attackers published litellm 1.107.0, a malicious release of the popular LLM proxy library, to PyPI and LiteLLM’s official Docker images. According to LiteLLM’s analysis, stolen GitHub CLI OAuth tokens from three engineers launched the private release workflow against a lookalike fork. The package drops a credential stealer that runs whenever Python starts and a backdoor that checks a GitHub file for further code. PyPI quarantined the version and the images were removed; how the tokens were first stolen is still unclear.

Unconfirmed: How the three GitHub CLI OAuth tokens were initially stolen is unknown, and no connection to the earlier TeamPCP-linked incident is established.

Key facts
  • litellm has about 23 million weekly downloads; affected release is 1.107.0.
  • Docker tags included v1.107.0, latest, main-latest and main-stable.
  • Payloads cook.pth, cook.py and sitecustomize.py persist after the package directory is deleted.
  • Exfiltration goes to www.litellm[.]dev and targets cloud, developer, AI-tool, browser and proxy secrets.
  • An earlier 2026 compromise of 1.82.7 and 1.82.8 was linked to TeamPCP; no link is established here.
Security assessment · AI, from the article

Impact: Steals cloud, developer, AI-tool, browser and proxy credentials

ProductAffectedFixed in
BerriAI litellm1.107.01.106.2
What to do
  1. 1.Manually remove cook.pth, cook.py and sitecustomize.py from site-packages.
  2. 2.Remove litellm 1.107.0 and revert to litellm 1.106.2.
  3. 3.From a clean machine, rotate exposed credentials and audit logs for the exposure window.
Analyze this story with Vani AIImpact, who is affected and what defenders should do, in a new chat

Indicators of compromise

auto-extracted · verify before use · export allAll
TypeIndicatorContext
domainlitellm.devpted with AES-256 and RSA-4096, to the lookalike domain www.litellm[.]dev/healthcheck . Backdoor ( sitecustomize.py ): Every hour,
sha2560aaf7ba346d41ef2f40390436738b001180d5a1c03aa093f121078a9969d9a1a75729aeb86d Malicious litellm macOS x86-64 .whl hash sha256:0aaf7ba346d41ef2f40390436738b001180d5a1c03aa093f121078a9969d9a1a Malicious litellm macOS ARM64 .whl hash sha256:462d0c9f67ca
sha25612d4ccc5c6f8d35cf4f2cf233b7f3e418e863e657d81ff5f5fc9e210005b94842fdcde054 Malicious litellm musllinux ARM64.whl hash sha256:12d4ccc5c6f8d35cf4f2cf233b7f3e418e863e657d81ff5f5fc9e210005b9484 Malicious litellm Windows x86-64 .whl hash
sha2561bbcb1443bc962e45f5aad53fe3d2c30600e0379715fb6c664cdd3ac126c339eecustomize.py Payload file dropped into sitepackages sha256:1bbcb1443bc962e45f5aad53fe3d2c30600e0379715fb6c664cdd3ac126c339e Infected litellm image index sha256:7598727f209d717041db37f
sha2561bbce8caa94f01c0aaa0b7ddae51553a8a32e1ce53d96d44fe583272aff7ad00d5e02ae6 Malicious litellm manylinux ARM64 .whl hash sha256:1bbce8caa94f01c0aaa0b7ddae51553a8a32e1ce53d96d44fe583272aff7ad00 Malicious litellm musllinux x86-64 .whl hash sha256:c2e32d0
sha256462d0c9f67ca522d9e0e4acd245e90b36ce0d5353604a66bb7c43a1ec4a6fcf778a9969d9a1a Malicious litellm macOS ARM64 .whl hash sha256:462d0c9f67ca522d9e0e4acd245e90b36ce0d5353604a66bb7c43a1ec4a6fcf7 Malicious litellm manylinux x86-64 .whl hash sha256:668169d
sha256668169d35b87bf51324080390b1b1c5ab12bd335cfe5f1a99e3689abd5e02ae64a6fcf7 Malicious litellm manylinux x86-64 .whl hash sha256:668169d35b87bf51324080390b1b1c5ab12bd335cfe5f1a99e3689abd5e02ae6 Malicious litellm manylinux ARM64 .whl hash sha256:1bbce8ca
sha2567598727f209d717041db37f7dca271d3a7d328a279890e0e1d55375729aeb86d15fb6c664cdd3ac126c339e Infected litellm image index sha256:7598727f209d717041db37f7dca271d3a7d328a279890e0e1d55375729aeb86d Malicious litellm macOS x86-64 .whl hash sha256:0aaf7ba346d
sha256c2e32d0ce2548aa662c18bf5001911a79fe5b30cf8357132275add02fdcde054ff7ad00 Malicious litellm musllinux x86-64 .whl hash sha256:c2e32d0ce2548aa662c18bf5001911a79fe5b30cf8357132275add02fdcde054 Malicious litellm musllinux ARM64.whl hash sha256:12d4ccc5c

From the original article

3 min read at securitylabs.datadoghq.com

Malicious releases of LiteLLM v1. 107. 0 were published to PyPI and container registries following multiple maintainer account compromises.

Datadog Security Labs

By @datadoghq · securitylabs.datadoghq.com

Read the full article at Datadog Security Labs

A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.