Cisco Talos finds CLOSEDQUORUM malware that lets AI models vote on attacks

Cisco Talos released CAIRN, an open-source hunt for AI-integrated malware, and disclosed CLOSEDQUORUM — a Windows implant that polls DeepSeek, Qwen, Mistral, and Gemini to pick its next move without a human operator.

Cisco Talos on Tuesday published an open-source toolkit for spotting malware that wires in AI chatbots, and with it disclosed what researchers say is the first publicly documented Windows implant to hand tactical command-and-control decisions to a panel of large language models, WIRED reported.

The framework is called CAIRN, short for Cognitive Artifact Intelligence Research Network. It hunts from metadata — embedded prompts, provider endpoints, orchestration clues, and related artifacts — rather than requiring analysts to download and run every sample. Talos detailed the release in a CAIRN introduction and a technical write-up of the find, The Closed Quorum.

CLOSEDQUORUM is a roughly 16.4MB Go-based Windows binary. After it lands, Talos said, it queries up to four commercial model providers — DeepSeek, Qwen, Mistral, and Google Gemini — tallies their votes, and executes the winning choice from a constrained menu that includes credential theft, process injection, and persistence. There is no dedicated attacker-run C2 server issuing live tasking in that loop, and no human operator selecting the next move. Stolen material is meant to leave through a Discord webhook. BleepingComputer separately covered the same disclosure.

Talos linked development artifacts to criminal-forum postings about carding dating to 2025, but said it could not confirm who built the malware or whether it has been used in real-world attacks. The public distribution build Talos examined carried placeholder API keys and a dummy webhook, so researchers did not observe a complete live end-to-end run from that sample.

Ryan Fetterman, who led CAIRN at Talos, told WIRED the point is to give defenders a shared way to track AI-integration fingerprints as the category grows beyond a handful of named families. Matt Olney, Talos’s senior director of threat intelligence, framed the shift as attackers moving from AI as a productivity aid to AI as operational scaffolding.

This brief covers the Tuesday CAIRN release and CLOSEDQUORUM disclosure. It is distinct from AI Tech Daily’s earlier coverage of open-weight models used in bug hunting and from product security wires on consumer AI agents.

Cisco Talos finds CLOSEDQUORUM malware that lets AI models vote on attacks

Sources