AI automated attack system BossAI targets Taiwan emergency response services

Security vendor Acronis disclosed on November 4 the AI-automated attack system BossAI: an attacker located in Hualien, Taiwan, used 12 AI agents to attack hundreds of organizations in 11 countries within 12 days, with Taiwan the most heavily attacked region and emergency response services the main target. The agents divided the work of infrastructure setup, vulnerability scanning, penetration testing and exploitation, and connected security tools such as Nmap, sqlmap and Metasploit through an MCP server. The emergency patient care records platform of the Taipei City Fire Department was breached, leaking electrocardiograms, ambulance footage and patient vital signs dating from 2018, and the 119 emergency dispatch system also had case records from 7 administrative districts obtained. Acronis noted that commercial models were used only to build and maintain the attack infrastructure, while the actual attacks were mainly carried out by modified versions of DeepSeek, GLM and Qwen with safety restrictions removed, among which Hermes was the primary attack agent, able to automatically perform asset reconnaissance, vulnerability analysis and exploitation.