Researchers uncover autonomous AI-powered malware with no human control mechanism

Cybersecurity researchers at Cisco Talos have discovered malware that operates with complete autonomy, making decisions by consulting multiple AI language models instead of relying on human command operators. The finding highlights an emerging threat landscape as attackers begin weaponizing artificial intelligence in more sophisticated ways.

New Framework Detects AI-Enhanced Threats#

To address the challenge of identifying malware that incorporates AI capabilities, Cisco Talos released CAIRN (Cognitive Artifact Intelligence Research Network), an open-source framework designed to classify and track such threats. The system works by identifying telltale markers—what researchers call "fingerprints"—that AI integration leaves behind in malware code, then uses these signatures to group related samples and identify patterns.

Ryan Fetterman, a security researcher who led development of CAIRN, explains that these traces provide a crucial window into attacker behavior: "The core idea is that AI integration has these vestiges, like fingerprints, that are left behind. That gives us a signal that we can use to track these samples, classify them, and look at what's happening."

The First Fully Autonomous Malware#

Using CAIRN, the research team uncovered CLOSEDQUORUM, a Windows-based malware that operates without any human oversight. The implant makes decisions by querying multiple AI models—DeepSeek, Qwen, Mistral, and Google Gemini—and then executing actions based on consensus responses from those systems.

The architecture creates a self-contained loop with no mechanism for external human control. Even if one AI service becomes unavailable, the malware continues polling the others, ensuring the system remains operational. The malware's design enables it to steal login credentials and cryptocurrency, with researchers tracing possible connections to cybercriminal forums discussing credit card fraud dating back to 2025.

A Larger Problem Than Expected#

When Fetterman initially investigated AI-integrated malware in summer 2024, he discovered only about nine documented examples. That limited picture prompted him to develop CAIRN as a more systematic detection tool. After several months of analysis using the new framework, he found approximately 20 additional instances of AI-enhanced malware, suggesting the actual scope is substantially broader than public reporting has indicated.