PoeLLM Malware Botnet Compromises More Than 3,400 Servers Using a Poem to Hide C2

· Original article ↗

Summary

Black Lotus Labs says PoeLLM has compromised more than 3,400 servers since April by targeting open-source AI services. The malware derives changing command-and-control addresses from words in a GitHub poem.

Key points

  • Black Lotus Labs reported that PoeLLM has compromised more than 3,400 servers since April.
  • The botnet targets exposed services including LiteLLM, Ollama, Gotenberg and Gitea; researchers first encountered its infrastructure while investigating an Ivanti Sentry vulnerability.
  • PoeLLM extracts four words from a GitHub poem and maps them through a hard-coded dictionary to construct its command-and-control address.
  • The threat actor can change the poem to rotate C2 addresses without updating the malware, making the infrastructure harder to identify through network monitoring.
  • Researchers observed exploit scanning and cryptocurrency mining. The malware also has remote-code-execution functionality that could enable abuse of AI models and further compromise.
  • The actor’s identity and activity beyond scanning and mining remain under investigation; researchers have not linked the campaign to other groups.

Article Details

Event Type
PoeLLM malware botnet targeting open-source AI services
Impact
Black Lotus Labs reported that more than 3,400 servers had been compromised since April. The botnet has been linked to exploit scanning and cryptocurrency mining; its malware can enable remote code execution and may allow abuse of AI models and compromised servers as proxies for further attacks.

People

Malware

Vendors

Products

Countries

Industries

Related Articles