Cisco Talos Discovers ClosedQuorum Malware Using AI Ensembles for Autonomous Cyberattacks

Cisco Talos disclosed "ClosedQuorum," a Go-based Windows malware that queries an ensemble of AI models—Google Gemini, DeepSeek, Qwen, and Mistral—to autonomously determine post-compromise attack actions. Using a voting mechanism across these LLMs, the malware decides on tactics such as stealing browser credentials, extracting cryptocurrency data, establishing persistence, and spreading laterally without requiring human operator commands. ClosedQuorum is the first documented malware to delegate tactical command-and-control (C2) decisions and post-exploitation workflows to an ensemble of commercial LLMs. This marks a significant evolution toward highly autonomous AI cyber threats that operate with unprecedented speed and adaptability while bypassing traditional human-driven C2 infrastructure. Built in Go, the implant gathers system reconnaissance data and submits it to the LLM ensemble, executing actions chosen through model consensus from a set of predefined strategies. The automated actions include browser credential theft, crypto wallet targeting, persistence mechanism setup, and lateral movement across local network environments.