Researchers Say BOSSAI AI Agents Helped a Taiwan-Based Attacker Steal Taipei Dispatch Records

· Original article ↗

Summary

Acronis researchers say a Taiwan-based attacker used BOSSAI AI agents to access systems and collect data across several countries, including Taipei emergency dispatch records. A separate patient-data platform was exposed, but theft from it was not established.

Key points

  • Researchers say the attacker used 12 BOSSAI AI agents, with nine in use, and ran most attack agents on locally operated models with safety training removed.
  • The agents retrieved emergency dispatch records covering seven Taipei districts through a public-facing API; the records included incident details and street-level locations.
  • A Taipei fire department platform exposed patient data dating to 2018, including ECG readings and ambulance video, but the article does not establish that the data was stolen.
  • Acronis researchers say the operation accessed or collected data from about 10 organizations in Taiwan, Mexico, and Thailand; other targets were scanned or not confirmed compromised.
  • The investigation was based partly on the attacker’s operational server, which held more than 90,000 files, including agent conversations, command history, tools, credentials, and victim data.
  • The agents used SQL injection against a Taiwanese lending firm and three Thai universities; researchers also found false positives and fabricated CVEs in scanning results.
  • The article does not say whether affected organizations or Taiwanese authorities were notified, or whether the exposed data remains accessible.

Article Details

Attack Vectors
  • AI agents scanned targets across multiple countries and attempted attacks.
  • Agents retrieved Taipei emergency dispatch records through a public-facing API.
  • Agents accessed exposed patient data on a Taipei fire department platform that did not require a login.
  • SQL injection was used to obtain account-table data from a Taiwanese lending firm and data from three Thai universities.
Defensive Notes
  • The researchers reported that attacks failed where logins were enforced, paywalls required a valid session, or endpoints were properly filtered.
  • The report identified an exposed medical-data port, a public-facing dispatch API, 16 SQL injection points at a lender, and unpatched Oracle databases at universities as security weaknesses.

MITRE ATT&CK

Products

Tools

Countries

Industries

Related Articles