Serving Delaware, Pennsylvania, Maryland, and New Jersey

A Poem Became a Malware Delivery Tool

Would you ever consider poetry a cybersecurity threat to your establishment? A recently uncovered poem-based malware delivery technique helped attackers control thousands of compromised servers. Learn more about it here.

How Poetry Was Used in a Cyberattack

Security researchers at Lumen’s Black Lotus Labs discovered a bizarre new campaign nicknamed “Canto Incognito,” and it has left the tech world completely stunned.

The attacker hid a command-and-control server address in a poem that malware on infected servers decoded. Researchers believe the attacker is likely based in Italy (or at least speaks Italian) because of their frequent use of Italian servers and the presence of Italian-language comments in the malware’s code.

The victims in this case aren’t your typical desktop users but businesses running cutting-edge AI-related services. Specifically, the malware targets internet-facing servers running software such as:

  • LiteLLM
  • Gotenberg
  • Ollama
  • Gitea
  • Ivanti Sentry

By exploiting vulnerabilities in these setups, the attacker has already quietly infiltrated over 3,400 servers worldwide, building an extensive network of compromised server infrastructure.

How Do These AI-Powered Cyberattacks Work?

This poem-based malware delivery exploit follows a very specific attack pattern.

  1. Hunt for Vulnerable AI Services

The attacker starts by scanning the internet for corporate servers running specific AI frameworks. After finding a weak spot, the hacker sneaks in and installs a custom piece of malware known as PoeLLM. This initial breach gives the attacker a quiet foothold right inside the business’s network infrastructure.

  1. Decode the Hidden Server Address

Once inside, instead of connecting to a standard, easily blocked web address to receive orders, the malware pulls off a unique trick. It silently visits a legitimate, ordinary-looking public GitHub repository to read a short, two-stanza poem.

Instead of relying on malicious prompt injection or conventional large language model exploitation, the malware uses a hidden dictionary to isolate four specific words in the text (like driver, diode, decryption, and string) and translate them into numbers. When strung together, these numbers form the numeric IP address of the attacker’s command-and-control server.

  1. Hijack Resources for Crypto Mining

With the server successfully hijacked and connected, the malware drops its payload. Because AI servers run on expensive, powerful GPUs and CPUs, they are prime targets for resource hijacking. The attacker installs heavy-duty cryptocurrency miners (like XMRig) to silently mine Monero tokens.

  1. Turn the Victim Into a Launchpad

The malware doesn’t just sit there and mine. It forces the infected computer to start scanning both internal and external networks for more vulnerable AI tools to exploit.

Each infected server becomes another scanner that searches for vulnerable systems, allowing the malware botnet infection to spread across more networks.

Don’t Let Your Servers Become Someone Else’s Tools

Does your company have AI infrastructure? These poem-based malware delivery techniques should warn businesses that threat actors are only getting more creative.

The Black Lotus Labs researchers warn that we may encounter similar attacks in the near future. Always evaluate the risks before adopting new tools and implement proactive safeguards.

Used with permission from Article Aggregator

To top