15Threats

AI ransomware: inside the exposed 'Phantom' setup

In October 2026, SOCRadar researchers described an exposed server used by affiliates of the INC and The Gentlemen ransomware groups. It contained a framework the operators called an "Autonomous Red Team Kill Chain Framework", or Phantom, in which AI models carried out the hands-on intrusion work while a human approved each step.

01What the researchers found

According to the reporting, the exposed directory logged 8,257 messages across 133 sessions. A human operator gave a victim-specific prompt and reviewed each step; the intrusions were executed through an AI coding tool using commercial models. AI-managed knowledge bases tracked victims, stolen credentials and campaign progress.

When a model refused a request, the operator reportedly rewrote it as an "authorized red team assessment" rather than switching models. Safety filters alone are not a reliable barrier against a determined attacker.

02The way in was familiar

Initial access came from exploiting internet-facing edge and remote-access systems at scale, including FortiClient EMS, SimpleHelp, cPanel and VPN gateways. Known, actively exploited vulnerabilities were part of the toolkit.

That is the important point: AI made the attackers faster and able to run more intrusions at once, but they still entered through unpatched, exposed systems. The basics decide whether the speed matters.

03What changes for defenders

Time between a public vulnerability and mass exploitation keeps shrinking. Treat actively exploited flaws on edge devices as emergencies measured in days, not monthly cycles. Know every internet-facing device you own—especially remote-access tools installed by IT providers.

Faster attackers also mean less time to notice lateral movement. Alert on new administrator accounts, unusual remote-management sessions and large outbound transfers, and make sure someone actually responds out of hours.

04Practical steps

Inventory edge devices and remote-access software, patch known exploited vulnerabilities first, require MFA on every remote entry point, and keep offline or immutable backups. Rehearse your response so decisions do not wait for a meeting.

This article reflects research published on October 23, 2026.

Sources & further reading

The Hacker News — Exposed infrastructure reveals how ransomware affiliates used AI agents

SecurityWeek — In Other News, October 23, 2026

Continue with MindSec

Build offensive and defensive skills with INE certifications

All publicationsNext publicationSafePay ransomware: the CISA and FBI warning
Skip to main content