Acronis’s Threat Research Unit (TRU) reports that a single operator in Taiwan ran a mesh of AI agents, which it calls BOSSAI, against hundreds of organizations in 11 countries. The confirmed harm that matters most to this audience is in emergency services: agents pulled dispatch case records for seven Taipei districts and found a fire-department patient-care platform exposing 2018 medical data without a login. Acronis’s summary calls this confirmed exfiltration of medical data, while its detailed account says only that the records were exposed; we use “exposed”. Acronis published its analysis on 4 November 2026.
What Acronis reports
Acronis says it found the operator’s server while it looked like a misconfigured development environment. It held more than 90,000 files, and TRU observed the mesh for 12 days.
- Dispatch records: agents accessed a city’s 119 emergency dispatch system (119 is Taiwan’s fire and ambulance number) and pulled case records from its API: incident times, response types, status fields and street-level locations for seven Taipei districts. Acronis stresses these are case records, not live ambulance tracking.
- Patient data: a Taipei fire-department patient-care records platform exposed ECG readings, ambulance video and vital signs from 2018, all reachable without credentials.
- Vendor access: a fire-dispatch technology vendor’s database admin panel accepted default root credentials, exposing 18 or more tables.
- Elsewhere: SQL injection data theft at a Taiwanese lender and at three Thai universities, and about 206,800 Mexican residential phone records assembled from public sources.
Acronis says the operator scanned or targeted hundreds of organizations, but the agents collected data or gained access at roughly 10 distinct targets across three countries (Taiwan, Mexico and Thailand), per Acronis.
How the AI was used
Acronis describes 12 agents, nine in use. Commercial models built infrastructure but, per Acronis, repeatedly refused offensive requests (with one exception: Grok repaired a broken proof-of-concept script that the operator then handed to an attack agent), so the attacking agents ran almost entirely on locally hosted “abliterated” models: DeepSeek, GLM and Qwen variants with safety training removed.
The agents also failed. Acronis reports that one scanning tool fabricated CVEs and over half of its high and critical findings were false positives or misclassified.
Sector impact
Acronis’s own conclusion is that the agents succeeded against weakly protected data: a legacy medical-data port open since 2018, dispatch data on a public-facing API, default credentials, unpatched Oracle databases. They did not get past enforced logins, session-gated content or properly filtered endpoints. Acronis attributes the operation to one person with high confidence, based on reused identities across his businesses and attack servers. Acronis says the motive is unclear, presumably money.
Gaps in the reporting
No evidence found as of 10 November 2026 in the Acronis analysis or The Hacker News coverage that the affected organizations or Taiwanese authorities were notified, or whether the exposed data is still reachable.
Recommended actions
Our assessment: we rate this high severity and tag it healthcare and critical-infrastructure because the confirmed harm touches emergency-response and patient data; that rating is ours, not Acronis’s. We also see nothing here that needs a new defense. It rewards basic hygiene done faster than a cheap scanner can find the gaps.
- Inventory internet-facing admin panels and APIs in emergency, health and public-sector systems, and remove default credentials.
- Close legacy data ports and put authentication in front of any record store, however old the data.
- Test public web applications for SQL injection and patch end-of-life databases.
- Alert on high-volume, systematic API requests. Our assessment: that is how an agent sweep would look server-side; Acronis does not describe signatures.