ClosedQuorum Windows malware lets four AI models vote on its next move after infection

Cisco Talos researchers have documented ClosedQuorum, a Go-based Windows malware that uses Google Gemini, DeepSeek, Qwen and Mistral models to decide what to do after it compromises a host, with no human operator issuing commands. BleepingComputer reports that the models vote using reconnaissance data. DeepSeek decides ties, followed in priority by Qwen, Mistral and Gemini. The choices are limited to a predefined set. 'Steal' dumps LSASS credentials, browser passwords from Chrome, Edge and Firefox, and cryptocurrency wallets. 'Inject' generates shellcode for process hollowing or Early Bird APC injection. 'Persist' runs a persistence module. A 'move' option for lateral movement exists, but the analyzed build has no handler for it. Stolen data goes to the operators through a Discord webhook, so after delivery the attack can run fully automatically. Help Net Security reports that Talos also released CAIRN, an open-source framework that finds AI-integrated malware using only file metadata, such as embedded prompts and provider API endpoints. It sorts findings into three confidence tiers. Defenders can use it to find such samples without running them.