Cisco Talos said it has identified a new malware strain called CLOSEDQUORUM that uses a panel of large language models to decide its next actions without human direction.. The malware is designed to steal login credentials and cryptocurrency wallets, and after deployment it delegates action selection to commercial LLMs and executes the resulting decision autonomously, Talos said in a blog post published Tuesday..
“After deployment, it delegates the selection of its next action to a panel of commercial large language models and executes the resulting decision, with the intent of harvesting user credentials and crypto wallets,” the researchers wrote. “It does not require continued commands from a human operator.”.
Talos said the malware has “fully autonomous command and control,” with the complete dynamic operation delegated to AI rather than a human operator.. The Cisco threat intelligence team disclosed the finding alongside a new research toolkit called Cognitive Artifact Intelligence Research Network, or CAIRN, which it developed to track and hunt AI-integrated malware..
CAIRN uses metadata to identify signs of AI inside malware, including prompt templates, provider endpoints, API keys, jailbreak terms and other embedded artifacts, Talos said in a separate blog post.. Talos said attackers building AI-integrated malware “unintentionally (and inevitably) leave behind markers of their own,” and described its hunting methodology as “fast and scalable.”.
The researchers said CAIRN was the tool used to discover CLOSEDQUORUM.. Talos said it does not have confirmation that CLOSEDQUORUM has been deployed in the wild.. The malware was named for its structure because it relies on a decision-making quorum of LLMs to act, Talos said.
It supports integration with DeepSeek, Qwen, Mistral and Google Gemini, though it can operate with fewer models.. To choose a course of action, CLOSEDQUORUM polls the LLMs, and no human input is required or supported, Talos said.. Featured image credit.
Tags: CiscoCybersecurity
