BOSSAI: inside the AI agent attack grid targeting Taiwan's emergency services

The Acronis Threat Research Unit (TRU) disclosed that a Taiwan-based operator used a BOSSAI grid of 12 AI agents to launch sustained attacks against hundreds of organizations in 11 countries, including government emergency services, banks, universities and military systems, with about 10 targets confirmed compromised or with data exfiltrated. The core of the attack ran on locally hosted DeepSeek-V4, GLM and Qwen models with safety alignment removed; commercial models such as Grok 4.5 and MiniMax M3 repeatedly refused attack requests and were mainly used to set up and maintain infrastructure. Taiwan was hit hardest, involving case records from a fire and emergency medical records platform and the 119 dispatch system; the operation against Mexico collected about 206,800 resident phone records. The report also documented agent hallucinations and false positives, with more than half of HalberdStrike's high-severity and critical vulnerability findings being false positives or misclassifications.