Threat watch / Attack
BOSSAI: a lone operator's AI agents take data from Taiwan's emergency services
Acronis researchers describe a single operator, based in Taiwan, who ran a mesh of 12 AI agents (nine active), mostly on local models with their safety training removed, against hundreds of organizations in eleven countries. They confirmed data theft or access at about ten, including Taipei fire and ambulance dispatch records, patient data left open without a login, and three Thai universities. The ways in were ordinary: SQL injection, default credentials and systems open without authentication.
- Sector
- Government & politics
- Way in
- Software flaw
- Product or vendor
- None named
- How it happened
- According to Acronis, the operator drove nine AI agents by hand, pasting output from one to the next. Attack work ran almost entirely on "abliterated" local models (versions of DeepSeek, GLM and Qwen), while Grok and MiniMax M3 refused attack requests. The break-ins used SQL injection (16 injection points at a lending firm, three Thai universities), default root credentials on a fire-dispatch vendor's database admin panel, and medical and dispatch data reachable without credentials.[1][2] Checked 10 Nov 2026
- What was stolen
- Taipei 119 fire and ambulance dispatch records across seven districts (incident times, response types, case status, street-level locations); patient data from a fire department platform, including ECG readings and ambulance video; a lending firm's account table; and data from three Thai universities, including student data.[1][2] Checked 10 Nov 2026
- Who was impacted
- About ten organizations in Taiwan, Mexico and Thailand where Acronis confirmed data theft or access, mainly emergency services, universities and a lender, plus about 15 more attacked without a confirmed break-in. Acronis does not name the victims or say whether they were notified.[1][2] Checked 10 Nov 2026
- Threat actor
- Not attributed. A single individual based in Hualien City, Taiwan, according to Acronis, which reports high confidence that one person ran the infrastructure. It links him through an identity reused across his legitimate businesses and attack servers, and thinks he was probably after money.
Timeline
- Acronis Threat Research Unit publishes its analysis of BOSSAI: a mesh of 12 AI agents (nine active) aimed at hundreds of organizations in eleven countries, with "confirmed data exfiltration of medical data and PII" at universities and emergency services providers. [1]
- The Hacker News reports that the scanning agent invented CVEs and that more than half of its high and critical findings were false or misclassified, and that the agents failed where logins were enforced or endpoints filtered. [2]