New ClosedQuorum Malware Uses AI Models to Automate Windows Attacks

Researchers at Cisco Talos have identified a new strain of Windows malware named ClosedQuorum that employs artificial intelligence to manage its post-compromise operations. The malware, which is written in the Go programming language, uses a panel of AI models to autonomously decide which tactical actions to take on an infected host without human intervention.

AI-Driven Decision Making

ClosedQuorum distinguishes itself by delegating command-and-control (C2) decisions to a voting system involving several large language models (LLMs). The malware queries models including Google Gemini, DeepSeek, Qwen, and Mistral to determine its next move based on gathered reconnaissance data. When the models reach a tie, DeepSeek is programmed to act as the tie-breaker, making the final decision based on the most appropriate option.

According to researchers, these AI models are restricted to a predefined set of malicious actions, including:

• Steal: Simultaneously performing LSASS credential dumping, browser credential theft (Chrome, Edge, and Firefox), and cryptocurrency wallet extraction (MetaMask, Exodus, and Ethereum).
• Inject: Generating shellcode and utilising process hollowing or Early Bird APC injection.
• Persist: Executing a dedicated persistence module to maintain access.
• Move: Attempting lateral movement, although the specific distribution analysed by researchers lacked a corresponding handler for this action.

Automating the Attack Chain

The malware automates the exfiltration of stolen data by passing information to operators via a Discord webhook. This design allows the entire attack chain to proceed automatically once the initial delivery is successful, removing the need for a human operator to issue commands in real-time.

Cisco Talos researchers described the discovery as an “architectural shift towards attack-chain automation.” While they have not yet confirmed active deployment in the wild, artifacts found within the binary connect the developer to criminal forum posts related to carding activities dating back to 2025.

The discovery was made using CAIRN, an open-source toolkit released by Cisco Talos to assist researchers in tracking and analysing AI-integrated malware. While the malware represents a significant evolution in autonomy, researchers noted it may face technical hurdles, such as hitting API rate limits or receiving malformed outputs from the commercial AI services it relies upon.