
Black Lotus Labs has uncovered PoeLLM, a malware campaign targeting exposed artificial intelligence services and other internet-facing applications.
Active since at least April 2026, the operation uses a poem stored on GitHub to help infected servers locate its command-and-control infrastructure.
The campaign mainly affects vulnerable deployments of LiteLLM, Ollama, Gotenberg and Gitea. Researchers also identified possible targeting of Ivanti Sentry. Victims are concentrated in the United States and Western Europe.
PoeLLM installs cryptocurrency miners and turns compromised machines into scanning and exploitation servers. This allows the attacker to earn money while using infected systems to find additional victims.
The supplied research reports thousands of affected servers, although its figures differ between sections: the summary lists more than 3,400 victims, while the detailed analysis describes almost 2,200. Daily activity reached roughly 800 servers at its peak.
PoeLLM Targets AI Servers
PoeLLM retrieves a poem titled “On the Nature of Connection” from a GitHub repository controlled by the account “ejejejdfbbebe.”
The repository is a fork of the Node.js website source, but researchers found no apparent connection between the malware and Node.js itself.
The attacker placed the poem inside a file named “dash.css.” Instead of containing a visible server address, the poem provides words that the malware converts into an IPv4 address.
Custom parsing code extracts four words or phrases from specific positions. A hard-coded dictionary maps these terms to numbers, which become the four parts of the C2 address.
Changing selected words redirects infected systems to another server without changing the decoding method. Researchers observed 11 poem updates after the first commit on April 13, 2026.
Several C2 addresses hosted exposed router administration interfaces, suggesting the attacker reused compromised network devices.
Black Lotus Labs assessed an association with an Italian-speaking operator, based partly on Italian-language comments and infrastructure connections.
The published IOC file lists 12 C2 addresses. Three remained active as of October 7, 2026: 92.119.164[.]50, 103.249.201[.]108 and 178.128.14[.]204.
Researchers linked the campaign to an ELF malware file named “libgcrypt.” It includes remote shell functionality, XMRig and Iron cryptocurrency miners, HTTP scanning capabilities and exploit deployment features.
Scanning focused heavily on ports 3000 and 4000, associated with Gotenberg and LiteLLM. Crafted POST requests instructed vulnerable servers to download malicious files from C2 infrastructure on port 81.
One analyzed sample targeted LiteLLM’s “/mcp-rest/test/connection” endpoint, which the research links to command injection vulnerability CVE-2026-42271. Infected systems subsequently contacted C2 ports 3778, 5001, 5002 or 9999, lumen said.
Indicators of Compromise
| IP Address | Role | Reported Activity Period | Status at Publication |
|---|---|---|---|
92.119.164[.]50 | C2 server | First seen July 23, 2026 | Active |
103.249.201[.]108 | C2 server | First seen August 13, 2026 | Active |
Note: IP addresses and domains are intentionally defanged (e.g., [.]) to prevent accidental resolution or hyperlinking. Re-fang only within controlled threat intelligence platforms such as MISP, VirusTotal, or your SIEM.
Join 16,000+ SOC teams using ANY.RUN to streamline threat investigations and reduce manual effort. Explore for your team
The post PoeLLM Malware Targets Exposed AI Servers and Uses GitHub Poem to Hide C2 appeared first on Cyber Security News.
Discover more from RSS Feeds Cloud
Subscribe to get the latest posts sent to your email.
