AI-Powered Malware 'ClosedQuorum' Automates Post-Compromise Actions on Windows

A newly discovered Go-based Windows malware called ClosedQuorum leverages multiple AI models, including Google Gemini, DeepSeek, Qwen, and Mistral, to autonomously decide its next steps after compromising a system. The malware uses a voting mechanism among the AI models to choose actions such as credential theft, code injection, or persistence, with DeepSeek breaking ties. Cisco Talos researchers note that while no active deployments have been confirmed, the malware represents a shift toward fully automated attack chains, though it may face issues with API limits or service outages.

ClosedQuorum's design removes the human from the post-exploitation loop entirely. The malware polls four AI services, each returning a preference among four predefined actions, with DeepSeek serving as the tiebreaker. Its "steal" routine targets LSASS memory, browser credential stores, and cryptocurrency wallets simultaneously, then exfiltrates everything through a Discord webhook, meaning only the initial delivery requires operator involvement.

Cisco Talos located the malware through its CAIRN tracking toolkit and found no evidence of active campaigns. The analyzed sample contained placeholder API keys and a dummy webhook, suggesting a development or testing phase. The developer's apparent ties to carding forums from 2025 hint at a financially motivated background, though the malware's reliance on commercial AI APIs introduces operational fragility through rate limits and service outages.