Malicious v1.107.0 Steals Cloud Credentials and AI API Keys
Blockchain security firm SlowMist has warned that LiteLLM, a widely used open-source artificial intelligence (AI) gateway, was compromised again in a software supply chain attack. A malicious version 1.107.0 was published to the Python Package Index (PyPI) and Docker registries on October 21, carrying malware built to steal cloud credentials and AI API keys and to run attacker code on infected machines.
LiteLLM is a Python library and proxy server that lets developers route requests to many large language model (LLM) providers through a single interface. Because it often sits between applications and paid model accounts, a single deployment can hold API keys for several providers at once, which makes it a valuable target.
On October 21, 2026, malicious LiteLLM v1.107.0 was uploaded to PyPI and Docker, initiating the supply‑chain breach.
From 11:39 to 13:23 UTC PyPI hosted the infected wheels, which were removed at 13:23 UTC.
Docker tags remained compromised until 14:20 UTC, after which they were repointed or removed, ending exposure.
SlowMist Flags Malicious LiteLLM 1.107.0 Release
In a Threat Intelligence alert posted on October 22, 2026, SlowMist said the compromised release affected PyPI packages and Docker images, including floating tags such as latest and main-stable. Floating tags are image labels that automatically point to the newest build, so users who pulled them during the exposure window could have received the malicious version without choosing it.
According to SlowMist, the injected code runs automatically when Python starts, steals cloud credentials, AI API keys, and SSH (Secure Shell) private keys, and enables remote code execution (RCE), meaning an attacker can run commands on the affected system.
How the LiteLLM Release Pipeline Was Breached
LiteLLM confirmed the breach in its official incident report published on October 21, 2026, which lists the incident as contained, with an investigation still in progress. The company said the malicious actions were carried out through the GitHub accounts of three engineers, apparently using stolen GitHub CLI (command-line interface) OAuth (Open Authorization) tokens.
The report describes a coordinated sequence. One account was used to override branch protection and change the private release workflow, a second account approved the production release, and a third was used to delete workflow records, which limited the logs available to investigators. LiteLLM said it has not yet identified how the attacker obtained the tokens and has revoked all tokens involved.
The compromised packages were built from an attacker-controlled lookalike repository named Berri-AI/litellm, rather than LiteLLM’s official BerriAI/litellm repository. LiteLLM said the official v1.107.0 source tag on GitHub did not contain the malicious files.
Exposure Window: Times and Affected Artifacts
All times below are in Coordinated Universal Time (UTC) on October 21, 2026, as listed in LiteLLM’s timeline:
- PyPI: All seven litellm==1.107.0 wheels were live from 11:39 to 13:23 UTC, when PyPI removed the release.
- Docker Hub: The v1.107.0, latest, main-latest, and main-stable tags pointed to the compromised image from 11:46 to 14:20 UTC, when they were removed or repointed to verified clean builds.
- GHCR (GitHub Container Registry): The same tags served compromised images from 11:46 UTC until they were found unavailable at 13:51 UTC.
LiteLLM said customers running its private enterprise release were not affected. The company added that its private registry is currently distributing an embargoed fix for a separate vulnerability scheduled for public disclosure on October 27.
What the Malware Does
LiteLLM’s report says affected installs drop files named cook.pth, cook.py, and sitecustomize.py into the Python site-packages folder. Python loads these files every time the interpreter starts, so the malware can run even in projects that never import LiteLLM directly, and even when a user runs pip to fix the problem.
The company said the payload includes a credential stealer and remote code loading tools. Its guidance identifies cloud identity roles, every provider key stored in the LiteLLM proxy database, LiteLLM master and virtual keys, database passwords and any secrets in the host environment as data that should be treated as exposed.
What LiteLLM Users Should Do Now
LiteLLM says users are affected if they installed an unpinned version through pip between 11:39 and 13:23 UTC, or ran one of the affected Docker tags between 11:46 and 14:20 UTC. Its remediation steps advise users to:
- Avoid running pip first, since any Python start triggers the malicious files.
- Delete cook.pth, cook.py, sitecustomize.py, and related cached files from every affected site-packages folder.
- Stop the background Python process linked to the malware.
- Reinstall pinned to litellm==1.106.2, or rebuild images and virtual environments from scratch.
- Assume credentials were stolen if version 1.107.0 ran on a host with outbound internet access, and rotate every key and secret that host could reach.
SlowMist gave similar advice, urging teams to avoid version 1.107.0 entirely, rebuild from clean images, and pin dependencies to 1.106.2. Further technical details are being tracked in LiteLLM’s GitHub issue on the incident.
Second LiteLLM Compromise in 2026
This is the second supply chain incident to hit LiteLLM this year. In March, malicious versions 1.82.7 and 1.82.8 were published, an attack that Crypto Times reported led to the theft of large volumes of data and credentials.
Both incidents targeted the release and distribution pipeline rather than the public source code, which means reviewing code on GitHub alone would not have revealed the threat. For crypto firms that use LLM gateways in trading bots, AI agents and internal tools, the episode is a reminder that a single unpinned dependency can expose exchange-linked infrastructure and cloud accounts. Teams that pulled LiteLLM through floating tags or unpinned installs on October 21 should treat those environments as compromised until they are rebuilt and all credentials are rotated.
Also Read: SBI Data Breach: 3,897 Customer Records on the Dark Web. Can Blockchain Protect India’s Banks?
Disclaimer: The information researched and reported by The Crypto Times is for informational purposes only and is not a substitute for professional financial advice. Investing in crypto assets involves significant risk due to market volatility. Always Do Your Own Research (DYOR) and consult with a qualified Financial Advisor before making any investment decisions.


