Researchers Expose AI-Driven Attack Mesh That Targeted Hundreds of Organizations in Eleven Countries

Acronis's Threat Research Unit observed a single operator in Taiwan running a multi-agent AI system called BOSSAI, after the operator left its files exposed on a misconfigured server. The mesh used twelve AI agents split into builder and attacker roles, with the operator manually relaying outputs between them. Commercial cloud models refused offensive requests, so the attacks relied largely on local models with safety alignment removed. Victims included emergency services and government-linked systems in Taiwan, a Taiwanese lender, Thai universities, and Mexican phone records, with most successful breaches exploiting long-standing, easily found weaknesses rather than sophisticated exploits.