AI malware just removed the human from the attack loop
Cisco Talos uncovers CLOSEDQUORUM, first 'LLM-as-C2' malware using multi-model judge panel to autonomously steal credentials and crypto wallets.
Cisco Talos's CAIRN research toolkit identified CLOSEDQUORUM, a 64-bit Go Windows binary described as the first 'LLM-as-C2' architecture that automates an entire attack phase without a human operator. A panel of LLMs (DeepSeek, Qwen, Mistral, Google Gemini) votes on next actions via constrained JSON decisions, with deterministic tie-breaking biased toward DeepSeek. It performs LSASS credential dumping, steals saved browser passwords from Chrome, Edge, and Firefox, and targets crypto wallets including MetaMask, Ethereum, and Exodus. Talos states there is no confirmation of CLOSEDQUORUM deployment in the wild, and simultaneously released CAIRN as an open-source toolkit for hunting AI-integrated malware via VirusTotal metadata.
- CLOSEDQUORUM delegates credential theft to an LLM judge panel with no human in the loop.
- Targets LSASS memory dumps, browser saved passwords, and MetaMask/Ethereum/Exodus wallets.
- Uses DeepSeek, Qwen, Mistral, and Gemini in sequence; ties resolve toward DeepSeek.
- No in-the-wild deployment confirmed by Talos.
- Talos released the CAIRN open-source toolkit for tracking AI-integrated malware families.
- ATT&CK
- T1003.001 OS Credential Dumping: LSASS MemoryT1555.003 Credentials from Password Stores: Credentials from Web BrowsersT1071.001 Application Layer Protocol: Web Protocols
- Organizations
- Cisco TalosCLOSEDQUORUMCAIRNDeepSeekGoogle Gemini
From the original article
4 min read at csoonline.com
Attackers using AI have greatly benefited when it comes to speed and scale, and now, says Cisco Talos, the technology has evolved to execute large portions of the attack chain entirely without human involvement. Researchers at the threat intelligence group have identified what they call the first “LLM-as-C2” architecture that can fully automate the command-and-control (C2) chain.
csoonline.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.