CLOSEDQUORUM Malware Uses AI Models to Automate Cyberattacks

Cisco Talos researchers identified a new Windows malware named CLOSEDQUORUM. The malicious software, written in Go, operates by consulting four AI platforms—Google Gemini, DeepSeek, Alibaba Qwen, and Mistral—to determine its next steps after infecting a system.

The malware functions autonomously, removing the need for constant human oversight. When the AI models vote on an action, DeepSeek holds tie-breaking priority, followed by Qwen, Mistral, and Gemini. This approach allows the attack to continue even when the human operator is inactive.

The virus targets sensitive information by dumping the LSASS process memory, stealing browser passwords, and extracting data from cryptocurrency wallets. It disguises its execution as a legitimate process and establishes persistence within the compromised system.

Each version of the malware is customized with unique API keys and a Discord webhook. Stolen data is exfiltrated to the operator's Discord channel using AES-256-GCM encryption, with keys rotating daily. Experts recommend behavioral analysis to detect suspicious AI platform access combined with unauthorized system modifications.