Anthropic Reports Disrupting Highly Automated Cyber Campaign Linked to China

Anthropic says it halted a cyber campaign believed to be connected to Chinese state actors, who used its AI system to conduct attacks with minimal manual oversight. The campaign targeted 30 organizations globally.
The hackers reportedly succeeded in breaching several institutions, accessing internal systems. They achieved this by instructing Claude Code to operate as though it were performing legitimate cybersecurity tasks.
The company said Claude handled the majority of steps independently, completing up to 90% of the operation. Anthropic described this level of automation as a turning point in AI misuse.
Claude’s mistakes, however, reduced the impact of the operation. It hallucinated findings, miscategorized data, and repeatedly mistook public information for sensitive intelligence.
Security analysts have raised contrasting views. Some call the incident a sign of rapidly advancing cyber capabilities, while others believe Anthropic may be exaggerating what amounts to complex automation rather than true AI autonomy.