CrowdStrike: four criminal groups broke into healthcare organizations and government systems, among others, by tricking Claude Code and DeepSeek
CrowdStrike described four previously unknown criminal groups that carried out attacks between May and August using AI agents, most often Claude Code and DeepSeek models. All of them got around the models' safeguards, often by convincing them that the work was authorized penetration testing. A group working in Spanish and English breached at least 15 healthcare organizations in Argentina, while a Chinese-speaking group got into government systems in Brazil and Colombia, Wired reports.