Cisco Talos Details AI-Directed Windows Malware and CAIRN Analysis Framework

Summary

Cisco Talos introduced CAIRN, a metadata-only framework for analyzing AI-integrated malware, and detailed CLOSEDQUORUM, a Windows implant that uses four LLMs to select actions. Static analysis found its decision loop, but researchers did not observe a live run.

Key points

CAIRN analyzes file metadata and cognitive artifacts, including embedded prompts and AI-provider endpoints, without downloading or executing malware.
The framework sorts findings into tiers covering AI-related strings, evidence of operational AI integration, and confirmed named malware families.
Talos says AI-specific evasion text has appeared in unrelated malware samples within 12 months of its first confirmed use in the wild.
CLOSEDQUORUM is a Windows implant that asks DeepSeek, Qwen, Mistral and Gemini to vote on actions such as stealing data, injecting code or establishing persistence.
The malware is written in Go and does not require continued human commands or a dedicated attacker-operated C2 server, according to Talos.
Static analysis confirmed the decision loop, but the examined sample had placeholder API keys and a dummy webhook, so researchers did not observe a complete live run.

Article Details

Attack Vectors: CLOSEDQUORUM is a Windows implant that queries commercial LLMs and uses their votes to select actions from a fixed menu: steal data, inject code, or establish persistence.
The implant is designed to harvest user credentials and crypto wallets. Researchers reported that the analyzed sample contained placeholder API keys and a dummy webhook address and was not observed operating against a live target.
The article reports AI-specific evasion text embedded to suppress analysis by AI-based sandbox scanners in unrelated malware samples.
Defensive Notes: CAIRN classifies AI-integrated malware from file metadata without downloading or executing samples, using cognitive artifacts such as embedded prompts, provider API endpoints, orchestration logic, and evasion text.
CAIRN separates findings into tiers for AI-related strings, behavioral evidence of operational AI integration, and confirmed named malware families.
The article describes CAIRN strategies including surface expansion, pivoting, and corpus analysis to identify unknown samples, related samples, and patterns.