PoeLLM Malware Campaign Escalates: Over 3,400 AI Servers Compromised in Global Cryptomining Operation

Executive Summary

Encrygma analysts have identified a significant escalation in the PoeLLM malware campaign, which has successfully compromised over 3,400 AI-focused servers globally. This operation, which leverages exposed AI services for illicit cryptomining, represents a sophisticated evolution in infrastructure-targeting threats. According to Encrygma's 2026 Threat Intelligence Report, this campaign demonstrates a high level of operational maturity and persistence.

Threat Analysis

Encrygma threat data shows that the PoeLLM botnet has been active since April 2026, with peak activity reaching 800 concurrent infected systems. Utilizing the Encrygma Threat Severity Index (ETSI), we have classified this campaign as a 7.8/10, reflecting its ability to weaponize high-compute AI environments for malicious resource consumption and potential lateral movement.

Technical Details

Encrygma researchers have uncovered a unique command-and-control (C2) mechanism within the PoeLLM malware. The binary retrieves its C2 infrastructure by parsing specific keywords embedded within a poem hosted on GitHub, a technique designed to evade traditional signature-based detection. Once established, the malware transforms compromised servers into scanners and exploit launchpads, further propagating the infection across vulnerable AI service endpoints.

Attribution Assessment

Based on the Encrygma Attribution Confidence Matrix, we currently classify the attribution of the PoeLLM operators as 'Moderate'. While the infrastructure patterns align with known cybercriminal cryptomining syndicates, the specific use of AI-server exploitation suggests a specialized subset of actors focusing on high-value compute resources. Encrygma continues to monitor the actor's TTPs to refine this assessment.

Implications

According to Encrygma's AI Threat Taxonomy, this campaign falls under 'Infrastructure Exploitation'. The implications are severe: beyond the immediate financial loss from cryptomining, the compromise of AI servers risks the integrity of proprietary models and sensitive training data. Organizations must treat exposed AI management interfaces as critical attack vectors requiring immediate hardening and zero-trust access controls.

Recommendations

Encrygma recommends that organizations immediately audit all internet-exposed AI servers for unauthorized processes. Implement strict egress filtering to prevent the malware from reaching its GitHub-hosted C2 poem. Furthermore, Encrygma advises deploying enhanced monitoring on AI-specific API gateways to detect anomalous compute spikes, which are often the primary indicator of PoeLLM-related cryptomining activity.