Poetry is the new AI security threat as PoeLLM malware infects 3K+ servers
Quoth the LLM, 'More and more'
A suspected Italian cybercriminal has infected over 3,000 servers with a malware known as PoeLLM since April. The malware targets enterprise AI infrastructure to mine cryptocurrency and expand its botnet. This is the first documented case of "adversarial poetry," a technique that turns harmful prompts into poems to trick large language models (LLMs) into bypassing safety guardrails.
The attackers may have used a poem as a "perfect vehicle for hiding an important message" that remains unnoticed by security professionals. PoeLLM malware has been active primarily in the US and Western Europe, scanning for vulnerable internet-facing versions of LiteLLM, Ollama, Gotenberg, and Gitea. The malware abuses these open-source tools and can turn compromised systems into vulnerability scanners and exploit servers.
The threat originates from an Italian-speaking criminal, named Canto Incognito, whose GitHub repository contains a poem titled "On the Nature of Connection." The poem contains hidden commands that the malware uses to discover new command-and-control (C2) servers. As enterprises increasingly use AI in their operations, the attack surface grows, making them more vulnerable to such threats.
Written by urgent.news from The Register's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
