AI-Voting Malware Needs No Human Operator

Cisco Talos has uncovered CLOSEDQUORUM, a Windows credential-stealing malware that uses a panel of four AI models — Google Gemini, DeepSeek, Qwen, and Mistral — to vote on its next attack move every five to fifteen minutes, with no command-and-control server required. Talos also released an open-source toolkit called CAIRN on GitHub to help hunt similar AI-driven malware. The malware steals credentials from memory, browsers, and crypto wallets, exfiltrating data via Discord webhooks. Talos links it to a credentials-as-a-service operation tied to criminal forums active since 2025. No victims have been identified yet, but researchers warn that AI is rapidly shifting from a malware add-on to its core decision-making engine.