Cisco Talos releases open-source CAIRN framework, identifies Windows malware CLOSEDQUORUM that queries multiple LLMs for autonomous command-and-control decisions

Cisco Talos released CAIRN, an open-source framework for detecting AI-related fingerprints in malware, and used it to identify CLOSEDQUORUM, a Windows malware sample that queries up to four LLMs (DeepSeek, Qwen, Mistral, Google Gemini) to reach a "consensus" on its next actions with no human oversight. Talos says CAIRN also surfaced roughly 20 previously undocumented AI-integrated malware samples, versus around nine known families before this. The malware is designed to steal credentials and cryptocurrency and has loose links to credit-card-fraud forum activity since 2025, but there is no confirmed real-world deployment or victim.

If accurate, this gives defenders a concrete new detection tool and a documented example of malware attempting autonomous LLM-driven decision-making, which matters for security teams building AI-aware detection rules and for enterprises assessing exposure to a new malware category. The practical impact today is limited: this is one sample with unconfirmed deployment, not an active campaign, so the immediate action for most organisations is awareness and monitoring rather than emergency response.

The framing of an "AI hive mind" with "no humans in sight" oversells a malware sample that queries commercial LLM APIs for decision support, which is a novel engineering choice but not evidence of emergent autonomous coordination. The extraction has no primary source links, meaning this is Wired's account of a Talos disclosure rather than the underlying technical report, so the ~20 additional samples and the "attackers operationalizing AI" trend claim cannot yet be independently checked against Talos's actual data.

1. 01Whether Cisco Talos publishes the actual technical report or repo (CAIRN on GitHub/GitHub-equivalent) with IOCs, code samples and detection rules for CLOSEDQUORUM, rather than just the Wired summary
2. 02Any confirmed real-world detection of CLOSEDQUORUM in the wild (endpoint telemetry, incident reports, victim disclosures) versus it remaining a researcher-discovered sample with no confirmed deployment
3. 03Whether other vendors (Microsoft, Google Mandiant, CrowdStrike) corroborate the ~20 additional AI-integrated malware families Talos claims to have found using CAIRN, and whether CAIRN sees independent adoption as a detection tool