Authorities in Taiwan are investigating a scheme in which hackers allegedly used artificial intelligence agents to steal information from emergency medical calls and sell it to funeral companies. According to the Investigation Bureau of the Ministry of Justice, the data allowed these companies to contact family members shortly after patients' deaths to offer funeral services.

The case was disclosed this Monday (16), after a police operation carried out on November 12 and 13. Three people are under investigation and two had preventive detention authorized by the courts.

The authorities say the attacks began in April 2026 and hit dispatch systems for ambulances and rescue teams of national and local agencies. Among the information obtained were records of out-of-hospital cardiac arrest incidents, known by the acronym OHCA.

The investigation led to searches at eight addresses in the Hualien, Kaohsiung and Taoyuan regions, with computers, cellphones and other digital evidence seized. According to the agency, the suspects used virtual private servers (VPS) abroad to operate an automated attack system called BOSSAI.

What the technical investigation revealed about the BOSSAI system

BOSSAI had already been identified by Acronis' threat research unit, which published a technical analysis on November 4, before the police operation was announced.

The company described an infrastructure made up of 12 AI agents, nine of which were active during the investigation. They performed different functions, including server maintenance, vulnerability identification, tool development and attack execution.

Despite the automation, the system did not work entirely on its own. Acronis identified an operator who coordinated the agents, manually transferring results and instructions between different tools.

During 12 days of observation, the researchers identified activities targeting organizations in 11 countries, including financial institutions, universities and public services. In Taiwan, the team documented the extraction of emergency call records containing times, incident types and locations. These records did not correspond to real-time tracking of ambulances.

The analysis also revealed important limitations of the technology. Commercial AI models repeatedly refused requests to carry out offensive activities without authorization. To circumvent these restrictions, the operator resorted to local models modified to remove safety mechanisms.

These models, however, had frequent failures. Some agents identified nonexistent vulnerabilities, classified unsuccessful attempts as successful attacks and produced incorrect reports, hindering the operator's own work.

The Acronis investigation documents international operations associated with the BOSSAI infrastructure, but does not prove that the three suspects identified by Taiwanese authorities took part in all these episodes. The police statement also does not establish that responsibility individually.

The criminal case remains under investigation. So far, the judicial measures disclosed are precautionary and do not represent convictions.

More from Radar