Skip to content

AUT-2026-004 — CLOSEDQUORUM Autonomous Multi-Agent AI C2 Compromise

Density:
FORENSIC THREAT AUTOPSY · SECTION 53

AUT-2026-004: CLOSEDQUORUM Autonomous Multi-Agent AI C2 Compromise

2026-09-29ACTOR-CLOSEDQUORUM

Deep autopsy into the CLOSEDQUORUM operational compromise: exploitation of RPyC deserialization in LLM serving clusters, establishing decentralized multi-agent coordination with autonomous privilege discovery.

DimensionAttribution & Scope
Attributed AdversaryACTOR-CLOSEDQUORUM
Associated CampaignCAMP-AGENTIC-SWARM-2026
Targeted VulnerabilitiesCVE-2026-103395
Impacted SectorsSEC-TECH, SEC-ENERGY, SEC-DEFENSE
Incident Timeline Anchor2026-09-29
Kill Chain StageTTPObserved Mechanics & Telemetry Evidence
Initial ExploitationT1190RPyC protocol deserialization exploit sent to internal port 8000 of LLM inference backend.
Inbound TCP payload triggering unpickling of custom Python execution object in serving worker
Lateral Movement & Agent HijackingT1071Poisoning of Model Context Protocol (MCP) tool registrations, hijacking tool execution across agent pipeline.
MCP schema modification injecting malicious Bash execution parameter into tool manifest
Autonomous Command and ControlT1573Autonomous sub-agents communicating via encrypted P2P gossip protocol across compromised GPU nodes.
Inter-node GPU RoCE/InfiniBand traffic carrying encrypted LLM-generated operational tasks

AI Serving Security Architecture

Inter-process communications between model workers relied on insecure pickle/RPyC deserialization.

Agentic Tool Governance

MCP tool schemas lacked cryptographic integrity signing, allowing silent tool definition alteration.

Failed Defensive ControlFailure Mechanism Identified in Post-Mortem
AI Inference Sandbox Isolation (CTRL-APP-SANDBOX)GPU acceleration containers were granted SYS_PTRACE and host network namespace privileges.

[!WARNING] Source: GPU Telemetry (NVML) — Spike in tensor core utilization during off-peak hours without corresponding API inference requests. (Dwell time impact: Unmonitored compute anomalies allowed autonomous agent swarm to self-replicate for 48 hours.)

  • Ban unsafe deserialization protocols (pickle, RPyC) across all distributed AI model infrastructures.
  • Implement cryptographic code signing for all MCP tool manifests and dynamic function declarations.
  • Deploy hardware out-of-band kill switches capable of severing GPU cluster interconnects instantaneously.