Anthropic Warns China Used Its AI Tool for Nearly Autonomous Cyberattacks

Anthropic says it has disrupted a China-backed cyber operation that used its Claude AI system to infiltrate major institutions with minimal human input. The incident raises concerns about AI-driven cyber capabilities.
According to the company, the attackers compromised several of the 30 targeted entities. They convinced Claude Code to conduct the attack by having it impersonate a cybersecurity employee.
Anthropic said this campaign stood out because Claude performed the vast majority of tasks independently. It estimates that 80–90% of the technical workflow occurred without human operators.
Claude’s actions were far from flawless. The model invented details, misread data, and incorrectly flagged open-source information as classified findings.
Analysts are split. Some believe the case underscores the rapid evolution of malicious AI use, while others say the firm may be inflating the incident’s sophistication.