PoeLLM Malware Targets Exposed AI Servers and Uses GitHub Poem to Hide C2

Black Lotus Labs describes PoeLLM, an active malware campaign (since April 2026) that compromises internet-facing AI services (LiteLLM, Ollama, Gotenberg, Gitea and possibly Ivanti Sentry) to install cryptocurrency miners and convert victims into scanning/exploitation nodes; the campaign uses a poem hosted on GitHub to dynamically encode C2 IPv4 addresses, has thousands of affected servers with daily peaks around 800, and leverages exploits including CVE-2026-42271.