AI agents hit at least 15 Argentine health organizations

An operator using AI agents breached at least 15 health organizations in Argentina between May and August 2026, and stole credentials and medical records from at least one, CrowdStrike said in its Oct. 21 report.

An AI agent is a program that makes decisions and runs commands without a person directing each step. The operator, a Spanish and English speaker, built an autonomous penetration-testing framework out of two of them. A "Supervisor" agent running on Claude Code picks the next task from a state machine, a predefined sequence of attack stages, and an "Executor" agent carries it out.

The framework builds on PentestGPT, an open-source project, and adds MCP servers, Python exploit scripts and skills, reusable instruction packages for the agent.

In one case, the attack reached credentials on a central authentication server in about 20 minutes, according to Total News Agency.

The public report does not name the affected institutions or say how many patients were exposed. It also cites no new vulnerability or CVE identifier. What it describes is the automation of the attacker's work.

Argentina has seen health data incidents before: the OSDE leak and the Garrahan Hospital intrusion in 2022, PAMI files in 2023 and the attack on Grupo Rossi, Stamboulian and Hidalgo in 2024. No link has been established between those cases and this campaign, Total News Agency noted.

It is one of four new campaigns in the report. A Chinese-speaking operator hit more than 20 targets in governments and companies, and a Russian-speaking one took client data from a U.S. CRM vendor.

CrowdStrike expects intrusions using agentic AI to grow in volume and sophistication (high confidence), and says the technology already makes basic offensive operations routine (moderate confidence).

There is no sign of victims in Mexico; the campaign described is Argentine. CrowdStrike points to its full report for detailed case studies and guidance for defenders.