Cisco Talos, the threat intelligence arm of Cisco, has introduced a new research toolkit named CAIRN (Cognitive Artifact Intelligence Research Network) designed to track and classify emerging AI-integrated malware. The development of CAIRN has already led to the discovery of a malware binary called "CLOSEDQUORUM," which exhibits fully autonomous command and control (C2) capabilities. This finding represents a shift in offensive cyber operations, where portions of the attack chain can be executed by AI systems without human operators.
CAIRN operates by identifying "cognitive artifacts" left behind by attackers integrating AI into their tools. These artifacts include prompt templates, provider endpoints, API keys, and jailbreak terms embedded within the malware. The framework's metadata-first architecture allows it to extract, relate, and classify these artifacts without requiring the download or execution of the underlying binary. CAIRN employs rule-based detection, semantic clustering, and relationship graph traversal across up to 24 acquisition filters, examining metadata such as extracted strings, sandbox behavior, and antivirus detection labels.
The "CLOSEDQUORUM" malware, identified through initial hunts with CAIRN, is described as the first publicly documented Windows malware of its kind to demonstrate fully autonomous C2. While Cisco Talos has not confirmed in-the-wild deployment of "CLOSEDQUORUM," the artifacts linked to the binary have been traced to developer postings on criminal forums related to carding, dating back to 2025. This malware serves as an example of how attackers can constrain the decision space for an attack phase, allowing AI models to provide reasoning and act independently.
The progression from AI as an optional feature in malware to fully autonomous, multi-model consensus orchestrators with no human operator occurred within a single calendar year, according to Talos's observations. This rapid escalation highlights an "autonomy escalation arc" in AI-integrated malware development. The integration of AI in cybercriminal workflows has been a growing concern, with previous reports from Cisco Talos and CrowdStrike indicating that cybercriminals use AI to accelerate vulnerability exploitation, bypass security measures, and compromise software supply chains. Attackers have also been observed using AI to generate destructive scripts and streamline attacks, as seen with groups like Qilin.
The ability of AI systems to execute attack phases without human involvement compounds the effects of speed and scale, as human operators are no longer the bottleneck. An AI system can continue operations when a human operator is not actively monitoring, offering continuous attack execution. Cisco Talos plans to share further findings from its CAIRN project in subsequent reports.