---
title: CrowdStrike reports AI agents conducting cyber attacks
url: https://www.elseif.net/crowdstrike-reports-ai-agents-conducting-cyber-attacks
published: 2026-10-24T21:05:03+00:00
language: en
section: Agents
source: https://3dnews.ru/1150306
organizations: CrowdStrike, PentestGPT, Claude Code
publisher: elseif
---

# CrowdStrike reports AI agents conducting cyber attacks

CrowdStrike Counter Adversary Operations detailed research on four previously undocumented attacks using AI agents. Threat actors increasingly employ AI not only to create tools but also to execute large scale cyber attacks where agents autonomously make decisions. In Argentina a Spanish and English speaking adversary compromised at least fifteen medical facilities and stole medical records of at least one victim using a two agent autonomous pentesting architecture. The Claude Code Supervisor agent selected attack stages while the Executor agent performed them. The attacker leveraged the open source PentestGPT project. In China a Mandarin speaking adversary breached government structures of two nations and attacked private companies in Brazil and the United States across biotechnology medical food retail technology and tourism sectors.

The attacker accessed Claude models through a Chinese proxy service and built two pentesting systems offering varying autonomy levels. Agents can operate independently or delegate tasks to specialized agents. An Indonesian hacker used skills from public GitHub repositories and periodically intervened to overcome model defenses or prompt Claude Code to continue attempts. Despite limited security knowledge the actor compromised several Indonesian government domains and private companies. A Russian speaking group stole corporate client data from an American CRM provider and compromised private companies in Russia Kyrgyzstan and the United States. They developed a seven stage workflow granting AI agents near full autonomy in selecting attack methods.

Agent AI transforms basic offensive cyber operations into a commodity. The volume and complexity of such intrusions will almost certainly rise due to improved AI models and open source tool proliferation. In the next twelve months low skill adversaries may begin using AI systems for campaigns previously reserved for organized crime or state linked entities.