Attackers Leak Their Own AI Hacking Tools, Revealing How Autonomous Cyber Campaigns Work

A new report from the CrowdStrike Counter Adversary Operations team describes four previously unreported threat actors who left their offensive AI agent frameworks exposed in open directories. The exposure gave researchers detailed insight into how these groups use AI agents, and shows how quickly adversaries are becoming more skilled with the technology. Between May and August 2026, CrowdStrike Intelligence documented several incidents in which attackers used AI not just to build tools, but to run agentic campaigns where AI agents make decisions and execute commands.

Two of the actors are described in the opening of the report. The first is a Spanish and English-speaking operator who has compromised at least 15 healthcare organisations in Argentina and stole credentials and medical records from at least one victim. This actor used a purpose-built autonomous penetration testing harness with two agents: a "Supervisor" that picks the next task from a predefined workflow of attack stages, and an "Executor" that carries it out. The framework is derived from the open-source PentestGPT project, but was rebuilt with extra components.

The second is a Chinese-speaking operator with more than 20 named global targets. This actor compromised government entities in two countries and targeted private sector businesses in Brazil and the U.S., across biotech, healthcare, food and beverage, retail, technology and travel. This is only the opening of the report, so further details on the other actors are not covered here.