According to the Financial Times, hackers allegedly linked to China used AI agents to execute an autonomous cyberattack against Taiwan's government systems. The threat was detected by Dream, an Israeli cybersecurity company.

Over four days in early July, as many as eight AI agents operated simultaneously, mapping 21 government systems, identifying vulnerabilities and automatically changing strategy when encountering roadblocks. The operation compromised at least 85 accounts and extracted more than 2,500 personnel records .

The Dream researchers said the tool, built on top of the open-source Hermes and OpenClaw frameworks, implemented machine learning cycles: sessions where agents searched for CVEs and techniques in databases and GitHub. The system also self-corrected when it made mistakes.

Following the government's target, the campaign expanded to Taiwan's nuclear safety agency and at least seven companies in the energy sector, including government supply chain providers.

Although Dream did not officially attribute the attack to the Chinese government, the attackers' internal communications were written in Simplified Chinese, while the stolen data was in Traditional Chinese — a clue that points with high probability to Chinese operators.