Malware Uses AI-Powered Vote to Decide Next Move

Cisco Systems Inc.'s Talos Threat Intelligence group has discovered a Windows credential stealer called CLOSEDQUORUM, which takes its orders from no command-and-control server. The malware uses four commercial large language models to decide on its next move: Google Gemini, DeepSeek, Qwen, and Mistral. These models are queried in sequence every five to 15 minutes, with the winning decision determining which of four modules will be executed - stealing credentials, injecting code, establishing persistence, or moving laterally.

CLOSEDQUORUM's tactics include stealing Windows credentials from process memory, saved passwords from Chrome, Edge, and Firefox, and wallet files from MetaMask, Exodus, and Ethereum keystores. The stolen data is encrypted and pushed out through Discord webhooks. Persistence comes from registry run keys, scheduled tasks, and Windows Management Instrumentation subscriptions.

According to Talos researcher Ryan Fetterman, CLOSEDQUORUM represents a shift in effort displacement, with 'expanding portions of the attack chain' now running without an operator. Human attackers are bound by 'attention, working hours, and cognitive load,' while the panel answers every five to 15 minutes, indefinitely.