Anthropic Says Claude AI Powered 90% of Chinese Espionage Campaign

A China-linked state-sponsored threat actor manipulated Claude Code to carry out cyberattacks on roughly 30 organisations worldwide, according to Anthropic. The AI-powered campaign, identified in September, involved abusing Claude Code to inspect victim environments, identify high‑value assets, and exfiltrate credentials with minimal human intervention.

Anthropic notes that the attackers used the AI to perform 80–90% of the campaign, requiring human input only sporadically at perhaps four to six critical decision points. Nearly 30 entities across the chemical manufacturing, financial, government, and technology sectors were targeted, though only a small number were compromised. The attackers posed as a cybersecurity firm employee to bypass guardrails and fragmented their attack into small tasks for the model, enabling rapid operation and data exfiltration.

Within 10 days of detection, Anthropic says it disrupted the activity by banning identified accounts and notifying the targeted organisations. The report highlights how agentic AI can extend the reach of threat actors and accelerate cyberespionage. 14 November 2025.