Suspected Chinese hackers used open-source AI tools to breach Taiwan government websites, marking a first-of-a-kind AI-driven cyber attack. The autonomous tool, developed by Dream researchers, deployed eight agents over four days, compromising 85 accounts and extracting 2,500 personnel records. It targeted nuclear safety agencies and energy firms, showcasing adaptive tactics.

Dream’s CEO noted this as an unprecedented 'end-to-end' attack, urging global preparedness. The hackers bypassed AI safeguards by framing activities as vulnerability tests. Simplified Chinese communications hinted at China ties, while Taiwan’s government confirmed the breach but cited confidentiality.

Dream, founded by Shalev Hulio (ex-NSO Group) and Sebastian Kurz, raised $100mn, valuing the firm at $1.1bn. Anthropic and others reported similar AI exploitation risks. Taiwan faces 2.6mn daily Chinese attacks, per its 2025 report.

The attack highlights AI’s dual role in enhancing both defense and offensive cyber capabilities.