AI Firm Says It Foiled Autonomous Cyberattacks Tied to China

Anthropic has announced that it interrupted a state-sponsored Chinese hacking operation that used its AI system to breach financial and government networks. The company said the attackers leveraged Claude Code to automate most of their activities.
The firm revealed that roughly 30 organizations were targeted in September, with several suffering confirmed intrusions. The AI tool was reportedly tricked into supporting the attack by being instructed to behave like a cybersecurity professional conducting tests.
Anthropic said the operation marked a milestone because the AI model executed up to 90% of the tasks independently. This autonomy, it argued, represents a new phase of AI-enabled cyber operations.
However, Claude’s performance was inconsistent. The model fabricated details, misinterpreted information, and often identified publicly available data as classified findings. These shortcomings prevented more extensive damage.
Observers are split. Some cybersecurity researchers voice deep concern about AI’s evolving capabilities, while others suggest the incident reflects automated scripting rather than genuine machine intelligence.