Anthropic disrupts Russian, Chinese AI campaigns targeting its Claude models

Anthropic disclosed on September 10 that it had thwarted numerous malicious activities involving its Claude AI models over the past eight months. These included suspected Russia-linked cyber espionage and Chinese attempts to replicate Claude's capabilities. The company's latest Threat Intelligence report highlighted the growing trend of cybercriminals and state-backed hackers leveraging AI to execute cyberattacks and orchestrate portions of the attacks.

Human oversight was often minimal, with AI systems serving as primary orchestrators. The misuse of AI extended beyond mere chatbot interactions; it involved multi-agent frameworks executing complex tasks. Anthropic reported disruptions from seven China-based labs, including Alibaba, Moonshot, DeepSeek, and Xiaomi. Among these, Anthropic identified the largest illicit distillation attack, allegedly orchestrated by a group linked to Alibaba.

This attack aimed to extract Claude's capabilities to enhance the Chinese firm's Qwen models. Alibaba was unresponsive to inquiries. Additionally, Anthropic observed over 151 million exchanges associated with Alibaba between May and July 2026, peaking at nearly 3 million daily from over 3,500 fraudulent accounts. The tech firm accused Moonshot and DeepSeek of using live customer conversations, including sensitive data, as training data for Claude.

A Russian threat group, attributed to the SVR foreign intelligence service, allegedly employed Claude to devise phishing, Wi-Fi hijacking, and WhatsApp takeover operations against Ukrainian government, military, and diplomatic targets. This group also utilized AI to evade security defenses by automatically modifying malware. Anthropic uncovered new threat categories, such as the misuse of Claude for developing software to design weapons and related systems.

The company detected and disrupted activity linked to affiliates of the ShinyHunters cybercrime collective, one of the most active cybercrime enterprises, in recent months. Jacob Klein, Anthropic's head of threat intelligence, noted that AI models have become more capable, increasing risks.