China launched a fully autonomous vulnerability hunting attack against Taiwan
The attack leveraged eight open-source AI models to hunt for new attack vectors
The attack hit Taiwan government accounts, personnel records, the nuclear safety agency, and more
A first-of-its-kind cyberattack using autonomous AI has been spotted attacking Taiwan, and it compromised 85 government accounts and stole over 2,500 personnel records before moving on to hit the country’s nuclear safety agency and at least seven energy companies.

The attack used eight open-source AI models to build a hacking program that was able to independently conduct reconnaissance and intrusion, and was able to chain vulnerabilities and change tactics whenever it was blocked.

The intrusion took place over the course of four days, and was first exposed by The Financial Times on August 12, 2026. The FT article covered research performed by Dream, an Israeli AI and cyberdefense company that first identified the breach.

The attack was first uncovered during routine monitoring of cyber criminal activity. Dream found a 160MB online archive of 1,395 files. Further examination of the files revealed that the attack relied on Hermes and OpenClaw – two open-source AI agents.

As is typical of attacks relying on AI models, the hackers had framed the context of the intrusion as a routine cyber readiness test in order to bypass the built-in guardrails of the AI models.

The attack used multiple agents to hunt for new vulnerabilities and access points across the internet, providing the tool with multiple attack paths to take if one failed to gain entry.

AI agents have been quickly integrated into the attacks of cybercriminal organizations and state-sponsored threat actors alike, enhancing their abilities to launch highly complex attacks at scale. “This must be the basic assumption of every government around the globe,” said Amir Becker, Dream’s chief strategy officer.

Dream did not tie the attack to any specific cybercriminal group, nor did it confirm the target of the attack, but said it had alerted a government in the “Asia-Pacific.” Documentation within the recovered archive contained Simplified Chinese, which is the official written language used in mainland China.

The archive also contained data collected from the targets, which was written in Traditional Chinese. This form of written Chinese is widely used in Taiwan, Hong Kong, and Macau.

Read More: World-first autonomous ‘end-to-end’ AI attack hits Taiwan government systems