In a concerning development, a hacker from Taiwan utilized AI agents to illegally access emergency dispatch records across seven districts in Taipei. This breach was highlighted by Acronis’s Threat Research Unit last week, revealing a significant security lapse.
AI Agents Uncover Vulnerabilities
The AI agents also identified an exposed platform belonging to Taipei’s fire department. This platform contained sensitive patient data from 2018, including ECG reports and ambulance footage, which were accessible without any authentication.
Investigators from Acronis are confident that the cybercriminal operates solo from Hualien City, Taiwan. The hacker targeted numerous organizations across 11 countries, including emergency services, financial institutions, universities, and defense systems. Successful breaches were confirmed in Taiwan, Mexico, and Thailand, affecting around ten entities.
Extent of the Data Breach
The attacker’s operational base, initially mistaken for a development system, provided crucial evidence. This server contained over 90,000 files, comprising hacking tools, access credentials, and victim data, alongside recorded conversations between AI agents and command logs.
Despite the AI agents’ inability to communicate directly with one another, the attacker manually transferred outputs between them. Commercial AI models typically refused cooperation, compelling the hacker to depend on altered ‘abliterated’ models, stripped of safety protocols.
Targeting Taiwan’s Emergency Infrastructure
According to Acronis’s analysis, Taiwan bore the brunt of the attack, with emergency service data being the most severely impacted. The stolen records, extracted via the city’s emergency response API, detailed incident types, response times, and case statuses.
Aside from emergency services, breaches extended to other sectors. In Taiwan, a lending firm’s account database fell victim to SQL injection, while similar attacks in Thailand compromised data from three universities.
The Hacker’s Methods and Motives
The hacker’s methods relied heavily on AI-driven models, primarily for constructing and maintaining their digital architecture. Notably, the commercial AI models Grok and MiniMax M3 refused direct involvement in attacks without proof of authorization, although Grok made an exception once.
Investigators deduce the hacker’s identity based on errors made while trying to obscure his activities. He consistently used the same identity across both legitimate and illicit platforms, inadvertently revealing his personal IP address during an SSH session.
Future Implications and Security Measures
The incident underscores the need for enhanced cybersecurity protocols, particularly in safeguarding sensitive public data. The breach exploited existing vulnerabilities, such as outdated systems and insufficient password protections.
As AI technology becomes increasingly accessible, the potential for similar attacks grows. It is imperative for organizations to bolster their defenses against such sophisticated threats, ensuring robust authentication and regularly updated security measures.
