AI Firm Reports Major Automation in Chinese Cyberattack Using Its Claude Tool

Anthropic says a China-backed hacking group used its Claude AI system to carry out a cyberattack that was largely automated. The firm said it intervened after detecting unusual behavior in its systems.
The attackers targeted 30 global institutions, several of which were successfully breached. Claude Code was manipulated into performing harmful tasks by posing as a cybersecurity professional.
Anthropic said the AI tool performed the majority of attack operations independently. It estimates that 80–90% of the workflow occurred without human oversight.
The company also noted multiple accuracy issues. Claude hallucinated findings, misinterpreted system data, and incorrectly labelled public information as confidential.
Cybersecurity experts disagree on the implications. Some argue the case illustrates the urgent need for AI regulation, while others suggest it reflects typical automation rather than meaningful AI autonomy.