TechCapsules

AI-summarised brief · reviewed before publication

Researchers at Acronis’s Threat Research Unit disclosed that a lone hacker operating from Hualien City, Taiwan, used a suite of locally‑run “abliterated” AI agents, dubbed BOSSAI, to exfiltrate emergency‑dispatch records from seven Taipei districts and patient data from a fire‑department platform. The attacker targeted over 100 organizations in at least 11 countries, successfully breaching roughly 10 entities in Taiwan, Mexico and Thailand, and harvesting data such as ECG readings, ambulance video, loan‑firm SQL tables, university records and 206,800 Mexican home‑phone listings. Evidence was recovered from the attacker’s own server, which housed more than 90,000 files, including agent conversations, command logs, tools, credentials and victim data. Commercial AI models largely refused to assist, forcing reliance on locally‑run models stripped of safety training. The operation was observed for 12 days, revealing manual hand‑offs between agents and numerous false‑positive findings.