Malware · Cloud Security · AI Security
Poem-based malware infects thousands of servers via GitHub
Attackers used a poem on GitHub to control malware that infected over 3,000 servers, mainly targeting AI-related services. The malware mines cryptocurrency and can spread to more systems.
Ross GrantUpdated 3 min read
The short version
Attackers have infected thousands of servers with malware that uses a poem on GitHub to get its instructions. The malware mainly targets AI-related services and mines cryptocurrency using the victim's resources. Changing the poem lets the attacker control the malware without updating it. Only one news outlet has reported this so far, so details may change.
By the numbers
- 3,000+servers reportedly infectedTechRadar
- 11poem updates on GitHub since first commitTechRadar
- 2,500victims in previous LiteLLM supply chain compromise (for comparison)TechRadar
What happened?
Attackers used a poem posted on GitHub to control malware that infected over 3,000 servers, mainly targeting AI-related services like LiteLLM and Ollama. The malware, called PoeLLM, mines cryptocurrency and can spread to more systems. Only TechRadar has reported this so far.
Attackers use poetry to control malware
A new malware campaign has infected more than 3,000 servers by using a poem posted on GitHub as a way to control its operations, according to TechRadar (opens in a new tab). The malware, called PoeLLM, targets internet-facing services such as LiteLLM and Ollama. These are tools often used in artificial intelligence (AI) projects. The attacker appears to be based in Italy, but this has not been confirmed by any official source.
How the malware works
Once PoeLLM infects a server, it connects to GitHub and downloads a poem that appears to be AI-generated. The malware then scans the poem for specific words, such as "driver", "diode", "decryption", and "tick". Each word is linked to a number. When combined, these numbers form an IP address (a unique number that identifies a computer on the internet). This address tells the malware where to get its next instructions.
The attacker can change the poem on GitHub at any time. By changing certain words, they can update the IP address without needing to update the malware itself. This technique allows the attacker to move their command server quickly and easily. According to The Register (as cited by TechRadar), the poem has been updated 11 times, with the latest change in September 2026.
What the malware does
Once connected to the attacker's server, PoeLLM usually downloads and runs a cryptocurrency miner called XMRig. This program uses the infected server's computing power and electricity to mine Monero, a type of digital currency. The attacker profits from this mining, while the server owner pays the cost in energy and lost performance.
PoeLLM can also scan the internet for more vulnerable systems, helping the attacker spread the malware further. This makes the campaign more effective and allows the attacker to build a larger network of infected machines, known as a botnet.
Scale and uniqueness of the attack
Researchers told The Register that this campaign, called "Canto Incognito", is unusual because it targets multiple AI-related services at once. Previous attacks, such as the LiteLLM supply chain compromise, focused on a single service and affected about 2,500 victims. In contrast, PoeLLM has infected over 3,000 devices by targeting several services at the same time.
What should you do?
If you run internet-facing AI services like LiteLLM or Ollama, check for signs of unusual activity or high CPU usage, which can indicate mining malware. Regularly update your software and restrict access to only those who need it. Monitor GitHub repositories linked to your projects for any suspicious changes. If you suspect an infection, disconnect the affected server from the network and seek professional help to remove the malware.
What is still unclear?
So far, only TechRadar has reported on this attack, and there is no confirmation from GitHub, the affected vendors, or law enforcement. The exact method used to infect the servers is not fully detailed, and the total impact may change as more information becomes available.
Claimed vs confirmed
- Unconfirmed
Attackers used a poem on GitHub to control malware infecting over 3,000 servers.
Reported by TechRadar, but not confirmed by primary sources or affected vendors.
- Unconfirmed
The attacker is of Italian origin or based in Italy.
TechRadar suggests this, but there is no official confirmation.
The impact
- Money
- Victims may face higher electricity costs and lost server performance due to cryptocurrency mining. If personal data is involved, regulators could impose fines under laws like GDPR.
- Reputation
- Organisations running affected AI services may lose trust if their systems are compromised. The creative use of poetry for malware control could draw attention to security gaps.
- Customers
- Customers using affected services could experience slowdowns or outages. Those running AI servers should check for infection and update security controls.
Why it matters
This attack shows how creative attackers can be in hiding their control systems. It affects thousands of servers running AI tools, which are increasingly used by businesses. Anyone running exposed AI services could be at risk.
How does it work?
The malware downloads a poem from GitHub and scans it for specific words. Each word is linked to a number, and together they form an IP address. The malware then contacts this address for instructions. Changing the poem updates the command server without changing the malware.
Key points
- Malware called PoeLLM infected over 3,000 servers using a poem on GitHub.
- The malware targets AI-related services and mines cryptocurrency.
- Attackers can update the malware's command server by changing the poem.
- Only one news outlet has reported this so far; details may change.
- Defenders should monitor for unusual activity and restrict access to AI services.
Sources
TechRadar
Reference
