A single attacker employed AI agents to access emergency dispatch records across seven Taipei districts, as revealed by researchers at Acronis's Threat Research Unit This article explores targeted taiwanese emergency. . Additionally, the agents discovered that the Taipei fire department's platform was vulnerable, allowing unauthorized access to patient data from 2018, including ECG readings and ambulance video footage, without requiring a login.

Consequently, the attacking agents relied almost entirely on "abliterated" models, which run locally with their safety training disabled.

The agents also targeted five other entities linked to Taiwan's emergency services, but the report does not specify whether these organizations or Taiwan's authorities were informed of the exposure or if the affected data can still be accessed. In Mexico, they gathered approximately 206,800 home phone records, including names and addresses, using government statistics, postal code lists, and telecom authority data.

Cursor, another coding agent, targeted a Taiwanese emergency-services site while utilizing one of the obliterated models, which were versions of DeepSeek, GLM, and Qwen. They used a single identity across their legitimate businesses and attack servers, including a GitHub profile, logo, SSH key name, and email address.

The agents discovered that the data was inadequately secured, including a medical data port that had been open since 2018, access to a public-facing API, 16 SQL injection vulnerabilities in a lender, and unpatched Oracle databases at universities.