Security Firm Discloses Hackers Used Open-Source AI Agent Framework to Breach Asian Governments and Taiwan Systems

Threat intelligence firm Vulone disclosed on October 18 that a China-linked, financially motivated threat actor launched widespread cyber intrusions between September 24 and October 14 using the open-source AI agent framework Cairn. Targeting government agencies and enterprises across Asian countries including Taiwan, Vietnam, Indonesia, and India, the campaign reportedly succeeded in 48 out of 66 intrusion activities. The attack architecture used a central dispatcher to manage Kali Linux container nodes and integrated the DeepSeek V4.1 Flash language model to facilitate reconnaissance, vulnerability exploitation, and post-exploitation collection. In Taiwan, reported targets included systems of the Legislative Yuan, Ministry of Justice, Investigation Bureau, China Airlines, and messaging provider Every8D; in response, Every8D operator Team-Pro Technology stated that its internal environment lacked the specific mail services and system configurations described in the report.

Key takeaways

• Vulone reported that attackers used the Cairn framework with Kali Linux container nodes and DeepSeek V4.1 Flash to autonomously execute reconnaissance, vulnerability exploitation, and post-compromise data gathering.
• The attack platform operated up to 116 concurrent containers, generating roughly 230,000 event records over 12 days and consuming around 36.7 billion tokens via rotated accounts to reduce costs.
• The report identified Taiwan as a primary target, citing CGI command injection at the Legislative Yuan, webshell deployment at the Ministry of Justice, and probes against China Airlines and Every8D.
• Regarding the alleged compromise, Team-Pro Technology stated that Every8D does not operate bulk email services and lacks the network systems and architecture described by Vulone.