ClosedQuorum Malware Uses Four AI Models to Choose Attack Steps

Cisco Talos researchers found Windows malware called ClosedQuorum that consults models from Google, DeepSeek, Alibaba, and Mistral before choosing its next action, according to Windows Report. Talos describes it as the first publicly documented Windows implant that delegates tactical command and control decisions to several AI models.

ClosedQuorum collects reconnaissance data from a compromised computer and sends it to Gemini by Google, models from DeepSeek, Qwen, and models from Mistral AI. Each model chooses from predefined actions, which can include stealing credentials and cryptocurrency wallets, injecting code into processes, establishing persistence, and moving across a network. The malware can send stolen data through a Discord webhook.

Researchers have not confirmed ClosedQuorum's use in real attacks. The analyzed sample contained placeholder API credentials and a dummy Discord webhook, indicating it may be experimental. Its operation can also be interrupted by API rate limits, service outages, or model responses that the malware cannot process.