Suspected China-linked hackers carried out what researchers call the first known autonomous AI cyber attack on a government. The target was Taiwan, according to a person familiar with the matter.
The attack took place over four days at the start of July. It was uncovered by Dream, an Israeli cybersecurity company.
Dream found that the hackers built a tool using open-source AI systems called Hermes and OpenClaw. These systems let AI models carry out tasks on their own, without constant human input.
How the Attack Worked
The tool ran up to eight AI agents at the same time. Each agent acted like a member of a hacking team.
Together, they mapped 21 government computer systems. They also researched weaknesses in those systems and changed their methods when blocked.
The attack compromised at least 85 government user accounts. Hackers pulled more than 2,500 personnel records from these accounts.
The operation later spread beyond the original targets. It reached Taiwan's nuclear safety agency and at least seven energy companies.
Dream's chief strategy officer, Amir Becker, said he had never seen an attack this automated hit a government before. He said governments should now assume they face constant cyber attacks.
Dream has not named the group behind the attack. Researchers found Simplified Chinese in the hackers' internal messages, which suggests a link to China.
The data taken from the target, however, was written in Traditional Chinese. That script is mainly used on government websites in Taiwan, Hong Kong, and Macau.
A spokesperson for Taiwan's Ministry of Digital Affairs declined to comment on the specific hack. The ministry said all incidents involving government agencies follow set reporting and response steps.
The spokesperson added that AI has changed the nature of cyber security threats. They said AI agents create two new problems: attacks become automated, and the AI agents themselves can become new weak points.
Chinese authorities did not respond to requests for comment on the attack.
Growing Cyber Threat to Taiwan
Researchers said the hackers got around the AI model's built-in safety rules. They did this by telling the system its hacking activity was an approved test of the network's defenses.
Dream said the most unusual part of the attack was how the tool worked. It kept ranking different attack paths based on new information it gathered.
When one method failed, the tool sent out another agent to search the internet for fresh information. It then built a new plan, much like a human hacker would.
Dream was founded in 2023 by Shalev Hulio and former Austrian chancellor Sebastian Kurz. Hulio also co-founded NSO Group, the surveillance company blacklisted by the US in 2021.
Dream is based in Tel Aviv and describes itself as a national cyberdefence company. It was valued at $3 billion in June after a funding round backed by Bain Capital.
Taiwan's National Security Bureau reported in January that the island faced an average of 2.6 million Chinese cyber attacks a day in 2025. That figure was up 6% from the year before.
China claims Taiwan as part of its own territory. Beijing has said it could take Taiwan by force if the island refuses to accept its control indefinitely.
