Book an hour

LiteLLM: a poisoned AI update

For about two hours on 21 October, anyone who installed the newest LiteLLM got software that steals passwords and keys.

Confirmed, withdrawn. SYNC AI staff writers, 24 October 2026.

The source

‘This version contained compiled malware which steals credentials and loads remote-code execution tools.’
LiteLLM, 21 October 2026

What happened

LiteLLM is a popular tool that connects apps to AI models. Version 1.107.0 was published on PyPI from about 11:39 to 13:23 UTC on 21 October, and its Docker images until about 14:20 UTC.

The release was built from a lookalike copy of LiteLLM's code. The malware took cloud passwords, SSH keys, AI service keys, and the secrets of AI coding tools on the machine.

LiteLLM pulled the release. Every earlier version is safe. The last clean version is 1.106.2.

Confirmed by LiteLLM, Wiz, and StepSecurity. Nobody has named the attacker.

Reported by one source, not yet confirmed

LiteLLM says the attacker took a GitHub login token from its logs on 10 October. Only LiteLLM reports this.

Wiz says stolen keys were in use about four and a half hours after they were taken. Only Wiz reports this.

Why it matters

AI keys run on your account and you pay the bill. A developer's machine holds the keys to everything that developer builds.

What to do

  1. If 1.107.0 was installed anywhere, rebuild that machine or container from scratch. LiteLLM warns not to run pip there first.

  2. Pin LiteLLM to 1.106.2 or earlier.

  3. Change every password and key the machine could reach.

  4. If someone else built your AI tools, ask them whether they use LiteLLM.

Sources

All emerging threats