Threat Advisory

ClosedQuorum Malware Uses AI to Execute Attack Phase Autonomously

Threat: Malware
Targeted Region: Global
Targeted Sector: Technology & IT
Criticality: High
[subscribe_to_unlock_form]

EXECUTIVE SUMMARY:

CLOSEDQUORUM, a Windows malware binary, represents the first publicly documented implant to apply AI models for tactical command and control. After deployment, it delegates the selection of its next action to a panel of commercial large language models (LLMs) and executes the resulting decision, with the intent of harvesting user credentials and crypto wallets without human operator involvement. The complete dynamic operation is delegated to the AI, collapsing the decision space into a constrained set of choices. CLOSEDQUORUM's architecture supports up to four LLM provider integrations, each queried in sequence, their independent verdicts tallied, and the binary acts based on their judgment.

The process involves querying four commercial LLM providers: DeepSeek, Qwen, Mistral, and Google Gemini, with a structured prompt containing host context or offensive capability language. The system prompt is constrained to a typed JSON schema representing a specific attack-decision language, requiring executable decisions only. Responses are deserialized into a Go struct, routing to capability modules of main.main: steal, inject, persist, move. The LLM must emit a valid JSON object matching a known type, with a decision field that maps to a specific capability. If all models fail, the fallback decision is consensus: a string with no corresponding capability handler, causing the loop to sleep and retry.[/subscribe_to_unlock_form]

EXECUTIVE SUMMARY:

CLOSEDQUORUM, a Windows malware binary, represents the first publicly documented implant to apply AI models for tactical command and control. After deployment, it delegates the selection of its next action to a panel of commercial large language models (LLMs) and executes the resulting decision, with the intent of harvesting user credentials and crypto wallets without human operator involvement. The complete dynamic operation is delegated to the AI, collapsing the decision space into a constrained set of choices. CLOSEDQUORUM's architecture supports up to four LLM provider integrations, each queried in sequence, their independent verdicts tallied, and the binary acts based on their judgment.

The process involves querying four commercial LLM providers: DeepSeek, Qwen, Mistral, and Google Gemini, with a structured prompt containing host context or offensive capability language. The system prompt is constrained to a typed JSON schema representing a specific attack-decision language, requiring executable decisions only. Responses are deserialized into a Go struct, routing to capability modules of main.main: steal, inject, persist, move. The LLM must emit a valid JSON object matching a known type, with a decision field that maps to a specific capability. If all models fail, the fallback decision is consensus: a string with no corresponding capability handler, causing the loop to sleep and retry.[emaillocker id="1283"]

The operating model of CLOSEDQUORUM appears to operate as an operator-configured service rather than malware deployed directly by its developer. The publicly observed distribution binary is an inert template, requiring customization for each operator through injected LLM API credentials and a Discord webhook. Stolen credentials arrive in the operator's Discord channel, AES-256-GCM encrypted with a daily-rotating key derived from the message timestamp. This is obfuscation, not true confidentiality separation between developer and operator.

THREAT PROFILE:

Tactic Technique Id Technique Sub-technique
Execution T1059.006 Command and Scripting Interpreter Python
Persistence T1543.003 Create or Modify System Process Windows Service
Defence Evasion T1036.005 Masquerading Match Legitimate Resource Name or Location
Credential access T1555.003 Credentials from Password Stores Credentials from Web Browsers
Collection T1005 Data from Local System -
Command and control T1071.001 Application Layer Protocol Web Protocols
Exfiltration T1567.002 Exfiltration Over Web Service Exfiltration to Cloud Storage

MBC MAPPING:

Objective Behavior ID Behavior
Command & Control B0030 C2 Communication
Discovery E1083 File and Directory Discovery
Exfiltration E1020 Automated Exfiltration
Anti-Static Analysis B0032 Executable Code Obfuscation
Anti-Static Analysis E1027 Obfuscated Files or Information
Anti-Behavioral Analysis B0003 Dynamic Analysis Evasion

REFERENCES:

The following reports contain further technical details:

[/emaillocker]
crossmenu