IncidentAI-enabled attacks
INC and The Gentlemen affiliates: exposed servers show ransomware intrusions run through OpenCode, commercial models and 32 MCP servers
SOCRadar's Threat Research Unit published on October 20, 2026, and updated on October 23, its analysis of seven exposed servers used by affiliates of the INC and The Gentlemen ransomware groups. It has moderate to high confidence that operators of both groups used the same environment, low confidence in a formal alliance, and assesses the most likely structure as a tooling and infrastructure provider serving several ransomware partners. The operators drove intrusions through the OpenCode coding agent with commercial models bought through OpenRouter, Phantom, an "Autonomous Red Team Kill Chain Framework" with 18 skill modules, a WormGPT persona prompt, 32 MCP servers covering reconnaissance to extortion, and a knowledge base kept across three iterations. The records hold 8,257 messages across 133 OpenCode sessions and more than 23,920 dollars of AI API spending. The agent asked for the operator's approval before each step; when models refused, the operators rewrote the request as an "Authorized Red Team Assessment Request" rather than switching models. The operators' own records, which SOCRadar did not verify, show exploitation of FortiClient EMS (CVE-2026-35616), SimpleHelp (CVE-2024-57727) and cPanel/WHM (CVE-2026-41940). Of 62 named organizations, 36 are in the United States: 4 were listed for extortion, 4 reached encryption or extortion, 14 had domain-level compromise, 28 a foothold and 12 only reconnaissance. A September 30 screenshot shows the INC encryptor running on a victim workstation, with 46 builds packed for six victims.
Replay the attack step by step in the Learning Center
How it happened
- Break in through exposed edge and remote-access systems such as FortiClient EMS, SimpleHelp and cPanel, then escalate to domain admin.
- Run each step through an OpenCode session with a commercial model, MCP servers and a shared knowledge base, with a human approving every action.
- When a model refuses, reword the request as an authorized red team assessment, then encrypt, steal finance, HR and legal files and extort.
Impact
Footholds at 28 of 62 named organizations, domain-level compromise at 14, and encryption or extortion at 4; the ransom outcomes are unknown.
The fix
Patch FortiClient EMS, SimpleHelp and cPanel/WHM, take management interfaces off the internet, and look for admin accounts and VPN scripts you did not create. Require MFA, IP allowlists and dedicated admin accounts on remote-access and RMM tools, including your MSP's, keep immutable backups and turn on EDR tamper protection and the vulnerable driver blocklist.
Lessons
- Criminal crews use the same coding agents and paid models as developers; a refusal is reworded, not respected.
- AI speeds up the work after the break-in, but the way in is still unpatched edge devices and reused passwords.
- An attacker's exposed server can reveal the whole operation, prompts and victim list included.