AUTO-UPDATED

This Windows Malware is Built to Let Up to Four AI Models Vote on Its Next Move

Cisco Talos researchers have identified CLOSEDQUORUM, a novel Windows malware that automates malicious decision-making by polling multiple commercial AI models to determine its next operational steps on infected systems.

Key Points

  • The malware uses a voting system involving DeepSeek, Qwen, Mistral, and Google Gemini to choose between stealing data, injecting code, or establishing persistence.
  • Stolen credentials, browser passwords, and cryptocurrency wallet data are exfiltrated to the attacker via Discord webhooks.
  • Persistence is maintained through Registry keys, scheduled tasks, and WMI event subscriptions disguised as legitimate Windows Update processes.
  • Talos discovered the threat using CAIRN, an open-source tool designed to detect malware that leverages AI services for command-and-control operations.
  • While the current public version is non-functional due to missing API keys, the code has been in development since at least June 2026.

Why it Matters

This development marks a significant shift in cyberattack methodology by decentralizing command-and-control logic through third-party AI services. It forces security teams to monitor for unusual AI-service traffic patterns rather than relying solely on traditional server-based threat detection.
Internet Published by info@thehackernews.com (The Hacker News)
Read original