Cisco Talos uncovers malware that polls four language models for orders
Cisco Talos released an open-source framework called CAIRN and used it to find malware that takes its instructions from four large language models, with no room for human input.
Researchers at Cisco Talos published an open-source framework on Monday for classifying and analysing malware that embeds AI components. They call it CAIRN, short for Cognitive Artifact Intelligence Research Network, after the stone stacks hikers build to mark a trail.
CAIRN scans sample metadata for traces of AI integration, tags each artifact with a unique ID, and groups samples by shared traits against everything else in its library. Ryan Fetterman, the Cisco Talos security researcher who led the work, told Wired that AI integration leaves behind vestiges much like fingerprints, giving defenders a signal to track samples and see what attackers are trying.
A closed decision loop
One sample CAIRN surfaced is a Windows implant named CLOSEDQUORUM. To decide its next move inside a target system, it queries DeepSeek, Qwen, Mistral and Google Gemini and acts on the consensus. If one service is unreachable, it still polls the rest. That redundancy makes the loop entirely closed, with no mechanism for human input.
Cisco Talos found links between the malware and credit card fraud forums dating back to 2025. It is built to steal login credentials and cryptocurrency. The researchers could not confirm who wrote it, or whether it has been used in real attacks.
More samples than the public record shows
In July 2025, Ukraine's CERT-UA warned about a phishing campaign using malware known as LAMEHUG, which pulled commands from the Qwen2.5-Coder-32B-Instruct model through a Hugging Face API. Fetterman told Wired that when he ran a retrospective this summer, he could find only about nine named families, some of them research proofs of concept. After a few months of working with CAIRN, he has identified roughly 20 more examples. He says the practice is still largely experimental for attackers, but the landscape is more complex and diverse than public reporting suggests.
Matt Olney, senior director of threat intelligence at Cisco Talos, told Wired AI is moving from a productivity tool to something operationalised, letting attackers run more campaigns and handle more machines.
Sources: Cisco Talos, introducing CAIRN, the CLOSEDQUORUM analysis, Wired's interviews with the researchers.
