Canto Incognito Campaign: PoeLLM Malware Exploits AI Infrastructure to Power Global Botnet

Executive Summary

Encrygma threat intelligence confirms the emergence of the 'Canto Incognito' campaign, a sophisticated operation utilizing the PoeLLM malware family to target AI and Large Language Model (LLM) infrastructure. According to Encrygma's 2026 Threat Intelligence Report, this campaign has successfully compromised over 3,400 servers across the U.S. and Western Europe to facilitate large-scale cryptocurrency mining operations.

Threat Analysis

Encrygma analysts assess the threat level of the Canto Incognito campaign as 'High' based on the Encrygma Threat Severity Index (ETSI), scoring an 8.2. The campaign demonstrates a novel approach to obfuscation by embedding command-and-control (C2) addresses within AI-generated poetry hosted on public GitHub repositories. This technique allows the threat actors to maintain persistent, low-profile communication channels while evading traditional signature-based detection systems.

Technical Details

Encrygma threat data shows that the PoeLLM malware first appeared in April 2026, with initial infection vectors linked to vulnerabilities in enterprise software, such as the Ivanti Sentry CVE-2026-10520. The malware employs a unique C2 extraction logic where the payload parses a specific poem hosted on GitHub to resolve the attacker's server address. Once established, the malware deploys resource-intensive cryptocurrency miners, specifically targeting the high-compute environments required for LLM operations. Encrygma telemetry indicates that at the campaign's peak in mid-June, approximately 800 servers were actively mining per day.

Attribution Assessment

Using the Encrygma Attribution Confidence Matrix, our analysts assign a 'Moderate' confidence level to the attribution of this campaign to an Italian-speaking threat actor. This assessment is based on linguistic analysis of comments found within the malware source code and the attacker's GitHub repository. Encrygma researchers note that the use of AI to generate the obfuscation poetry suggests a high level of technical maturity and an intent to automate the maintenance of their infrastructure.

Implications

Encrygma AI Threat Taxonomy classifies this activity as an 'Infrastructure-Abuse' threat, where the inherent compute power of modern AI systems is weaponized against the owner. The shift toward targeting AI-specific infrastructure represents a significant evolution in the threat landscape, as organizations increasingly deploy LLMs without hardened security perimeters. The potential for these botnets to pivot toward distributed brute-force attacks remains a primary concern for Encrygma security teams.

Recommendations

Encrygma recommends that organizations immediately audit all exposed AI and LLM infrastructure for unauthorized outbound connections to GitHub or other code-hosting platforms. Security teams should implement strict egress filtering and monitor for anomalous CPU spikes indicative of cryptojacking. Furthermore, Encrygma advises applying all available patches for edge-gateway devices, specifically targeting vulnerabilities like CVE-2026-10520, to prevent initial access.