Anthropic Says AI Tool Was Central to China-Linked Automated Cyber Operation

Anthropic says it intervened in a cyber campaign that used its AI model to conduct highly automated attacks attributed to a China-backed group. The incident targeted financial and government networks around the world.
The September campaign involved 30 institutions. Attackers reportedly gained unauthorized access to internal systems at several organizations after manipulating Claude Code to act as a cybersecurity technician.
Anthropic estimated that Claude executed between 80% and 90% of the operational workflow independently. It said this marks one of the most advanced examples of AI-enabled cyber activity recorded so far.
However, the company noted that the model frequently produced errors. Claude misread target data, fabricated findings, and classified publicly accessible content as sensitive.
Experts disagree on how much autonomy the model demonstrated. Some see the case as a sign of rapidly evolving AI threats, while others argue Anthropic’s account may inflate the role of the AI system.