A cryptomining botnet has compromised more than 3,400 servers, many of them internet-facing LiteLLM and Ollama deployments, and hides its command server inside a poem.
BleepingComputer reports that Lumen's Black Lotus Labs has tracked a botnet called PoeLLM to more than 3,400 compromised servers since April, with up to 800 active on a single day across the US and Western Europe. Many victims run self-hosted AI tooling such as LiteLLM and Ollama, alongside Gotenberg and Gitea, and infected hosts scan for new targets and try to exploit CVE-2026-42271 in LiteLLM's MCP server test endpoints.
The malware installs XMRig and other cryptominers plus a remote shell, and locates its command-and-control server by decoding four words from a poem stored in a GitHub-hosted file, letting the operator move infrastructure simply by editing the verse.
Why it matters: Self-hosted AI gateways and model servers are now routine attack targets, so keep LiteLLM and Ollama patched, keep them off the public internet or behind trusted-IP allowlists, and check logs against the published indicators.