ClosedQuorum malware hands Windows attack decisions to AI models
The implant’s AI-led automation could speed up attacks and let them unfold at any time after infection.
Cisco Talos has detailed a new Go-based Windows malware, ClosedQuorum, that uses multiple LLM APIs (Google Gemini, DeepSeek, Qwen, and Mistral) to autonomously decide post-compromise actions. Instead of waiting for C2 commands, the implant gathers host data and lets the models vote on a limited action set such as credential theft, shellcode injection, persistence, or (theoretically) lateral movement. When the models disagree, DeepSeek’s choice wins ties, giving it effective priority over Qwen, Mistral, and Gemini in operational decisions. Stolen credentials and wallet data are exfiltrated via a Discord webhook, enabling almost fully automated attacks once the malware is delivered. Talos calls ClosedQuorum the first publicly documented Windows implant to hand tactical C2 decisions to an AI panel and warns this is an architectural shift that could increase speed and scale of future attack automation, despite current fragility from API limits and malformed outputs.
Why it matters
ClosedQuorum shifts parts of command-and-control from humans to a panel of commercial AI models, allowing the malware to decide on theft, code injection or persistence based on host data without operator input. That change means once a machine is compromised, credential and wallet theft and shellcode injection can proceed autonomously at any time, potentially increasing the speed and scale of attacks even as they inherit the fragility of external AI APIs and their limits.
Signal or noise?
Does this story matter, or is it hype? Decide before you see what everyone else thinks.