Cybersecurity Firm Discloses BossAI Attack System Targeting Taiwan Emergency Services

Cybersecurity firm Acronis disclosed an automated AI attack system named BossAI on November 4. According to Acronis, a threat actor located in Hualien City, Taiwan, developed an architecture of 12 AI agents that targeted hundreds of organizations across 11 countries within 12 days, spanning government emergency services, financial institutions, universities, military systems, and e-commerce platforms. Taiwan was the primary and most heavily targeted region, with emergency services being the primary focus. Individual agents handled infrastructure setup, vulnerability scanning, penetration testing, and exploitation, interfacing via an MCP server with tools including Nmap, sqlmap, Metasploit, Hydra, and Ghidra. Commercial models were used for infrastructure setup, while direct attack tasks were assigned to uncensored models—such as modified DeepSeek, GLM, and Qwen—with a primary agent named Hermes automating reconnaissance, analysis, and exploitation.

Key takeaways

• BossAI incorporates 12 AI agents coordinated via an MCP server with standard tools, using modified open-weight models for direct exploitation.
• Taipei's emergency care platform was breached, exposing medical records since 2018, alongside 119 dispatch incident logs across seven districts.
• The attacker identified hard-coded AES-256 keys in emergency management frontend JavaScript, risking session cookie and admin credential exposure.
• A Taiwanese lending service suffered SQL injection exposing database tables, while a mobile payment firm experienced authentication bypass due to flawed CVV verification.