LiteLLM disclosed on October 21, 2026 that its release pipeline was compromised in the early morning, pushing a malicious v1.107.0 build to PyPI and Docker Hub containing compiled malware designed to steal credentials and load remote-code execution tools. The company says the incident is contained and its investigation is in progress.
According to the incident post, the malicious litellm==1.107.0 package was live on PyPI for roughly two hours, from 11:39 UTC to 13:23 UTC, spanning all seven wheels before being pulled. On Docker Hub, the tags v1.107.0, latest, main-latest and main-stable pointed to a compromised image index from 11:46 UTC to 14:20 UTC; the v1.107.0 image was removed and the moving tags were repointed to secure sources.
LiteLLM said all previous versions are safe and that enterprise customers running the private enterprise release were not impacted. In a 20:30 UTC update, the company said it believes it has identified the attack chain, noting the attacker had to compromise multiple independent security controls to succeed.
The exposure window matters because LiteLLM sits in the hot path of many AI stacks: the open-source proxy routes requests to dozens of model providers and typically holds those providers' API keys in environment variables or configuration, exactly the class of secret a credential-stealing payload is built to harvest.
Any team that installed litellm==1.107.0 from PyPI or pulled the affected Docker tags during the window should treat host and workload credentials as compromised: rotate model provider keys, cloud credentials and any secrets present in the runtime environment, and audit egress traffic from affected containers for the exposure period.
The incident is the latest supply-chain attack on AI tooling, and it lands as enterprises standardize on gateway layers that centralize model access. LiteLLM says impacted artifacts have been pulled and pointed users to its disclosure for remediation guidance as the investigation continues.