Skip to content
NewA calmer, faster ZeroHour
Malwarecritical

Compromised LiteLLM 1.107.0 steals credentials via Python startup

Original title: Inside the LiteLLM 1.107.0 supply chain compromise: Credential theft through Python startup hooks

Microsoft Security Blog ()11 min read2 · 1 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

Starting Python with the package installed can run PRIVCOLLECT without importing LiteLLM; removing the package does not recover credentials.

On October 21, 2026, a compromised LiteLLM 1.107.0 release reached PyPI and official container images. Malicious cook.pth, cook.py, and sitecustomize.py files executed during normal Python startup, so an application did not need to import LiteLLM. Microsoft says the collector targeted developer, cloud, browser, SSH, environment, and LiteLLM proxy credentials, then encrypted and sent them out. Maintainers say stolen GitHub credentials redirected publishing to Berri-AI/litellm; Microsoft has not confirmed the initial access.

Unconfirmed: Microsoft has not independently determined the initial access or how the upstream credentials were obtained.

Key facts
  • Startup files were packaged under litellm-1.107.0.data/purelib/ and installed into the library directory.
  • cook.pth imports cook.py, which decodes an embedded program named PRIVCOLLECT.
  • A set PRIVCOLLECT_ACTIVE variable skips duplicate collection in child processes.
  • After HTTP 200, the collector tries to write /tmp/.pc_once_72129118c59f5636c0bb.
  • A Unix path could fetch more Python code; Microsoft did not confirm that activity.
Security assessment · AI, from the article

Impact: Theft of developer, cloud, browser, SSH, and LiteLLM credentials

ProductAffected
BerriAI LiteLLM1.107.0
What to do
  1. 1.Do not run Python or pip in a suspect environment to check for the package.
  2. 2.Treat package removal as insufficient; clean the environment and surviving processes.
  3. 3.Rotate credentials accessible to the compromised identity.
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
domainwww.litellm.devenvelope metadata. It then attempts an HTTP POST to hxxp://www[.]litellm[.]dev/healthcheck , using application/octet-stream , a fixed
domainwww.litellm.dev.exampleampaign domain. The query rejects suffix lookalikes such as www[.]litellm[.]dev[.]example; a match still needs process and workload context. De
sha25603ac2a8b752605d1d7710979c17ff18b03dd978203ba23ba97fd8ce3d6f897ea6f96a37eb86cf1ee2310d9be93139cc4ebef3ef9e2efe9531390be25", "03ac2a8b752605d1d7710979c17ff18b03dd978203ba23ba97fd8ce3d6f897ea", "34bef04fe88ac9d0aae0475630b1738681d15b94f233ac8aa46eacf5
sha25634bef04fe88ac9d0aae0475630b1738681d15b94f233ac8aa46eacf5968305fa752605d1d7710979c17ff18b03dd978203ba23ba97fd8ce3d6f897ea", "34bef04fe88ac9d0aae0475630b1738681d15b94f233ac8aa46eacf5968305fa", "b6122440d0d9cb1eb1ade9933f49214da8c31288de7710c076c2680d
sha2569dd0a3926f96a37eb86cf1ee2310d9be93139cc4ebef3ef9e2efe9531390be25an equivalent match. DeviceFileEvents | where SHA256 in~ ( "9dd0a3926f96a37eb86cf1ee2310d9be93139cc4ebef3ef9e2efe9531390be25", "03ac2a8b752605d1d7710979c17ff18b03dd978203ba23ba97fd8ce3
sha256b6122440d0d9cb1eb1ade9933f49214da8c31288de7710c076c2680dfa528465e88ac9d0aae0475630b1738681d15b94f233ac8aa46eacf5968305fa", "b6122440d0d9cb1eb1ade9933f49214da8c31288de7710c076c2680dfa528465") | project Timestamp, DeviceId, DeviceName, ActionType, Fi
urlhttp://www.litellm.dev/healthcheckaintext envelope metadata. It then attempts an HTTP POST to hxxp://www[.]litellm[.]dev/healthcheck , using application/octet-stream , a fixed browser-like

From the original article

11 min read at microsoft.com

Microsoft Threat Intelligence analyzes the LiteLLM 1. 107. 0 compromise, including Python startup hooks, credential theft, remote code loading, detections, and mitigations.

Microsoft Security Blog

microsoft.com

Read the full article at Microsoft Security Blog

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.