Detect Deepfakesby Resemble AI
Agentic AI attack

62 organizations in industrial agentic AI attack — Oct 2026

Ransomware affiliates utilized the Phantom AI framework to autonomously execute intrusions and exfiltrate data from 62 organizations across multiple critical.

Reported date
Oct 24, 2026
Target
62 organizations in industrial, manufacturing, energy, healthcare, and pharmaceutical sectors
Agent type
Other AI agent
Agent role
Used by the attacker

The exact incident date was not established. This entry is dated by its source report.

Updated Oct 24, 2026 · 2 min read

What happened

Cybersecurity researchers have identified an AI-driven intrusion ecosystem linked to the INC and The Gentlemen ransomware groups. The operation centers on a framework called Phantom, an autonomous red team tool used to exploit edge devices and facilitate data extortion. According to SOCRadar, the threat actors targeted 62 organizations across the industrial, manufacturing, energy, healthcare, and pharmaceutical sectors, with over half of the victims located in the United States.

The campaign utilizes an AI-executed, human-approved operating structure. Human operators provide victim-specific prompts, while the Phantom framework executes the intrusions using commercial models hosted on OpenRouter. The framework employs an agent that communicates with an external server to retrieve tasks, execute commands, and return outputs. It features 18 skill modules for reconnaissance, binary reverse engineering, and EDR fingerprinting. The attackers also utilized WormGPT-MCP, a set of 32 Model Context Protocol servers, to connect the AI system to various offensive modules.

Initial access was achieved by exploiting vulnerabilities in public-facing edge devices, including FortiClient EMS, SimpleHelp, cPanel, and various VPN gateways. Once inside, the operators followed a recurring lifecycle: enumerating the domain, escalating privileges, harvesting credentials, and exfiltrating data via reverse-SOCKS tunnels. The attackers used AI-managed knowledge bases to track victim telemetry and campaign progress. If models refused to act on prompts, operators reportedly rewrote requests to frame them as authorized red team assessments.

Beyond the Phantom framework, the toolkit included Ghost C2 polymorphic implants for persistence and RevSocks for network pivoting. The actors also employed a BYOVD (Bring Your Own Vulnerable Driver) attack using a Zemana driver to impair security products. Following data exfiltration, the actors analyzed stolen financials to determine ransom amounts and pressured victims through scripted vishing calls and threats to notify regulators. While four organizations were listed on ransomware data leak sites, the final outcomes of the ransom demands remain unknown, as no payment records have been recovered.

Evidence in the reporting

Incident evidence
A total of 62 organizations are estimated to have been targeted
Agent involvement
employs an agent that communicates with an external server to retrieve tasks

Sources