Anthropic disrupts Russian, Chinese AI campaigns targeting Claude models

Anthropic's Claude AI models have thwarted several alleged malicious uses over the past eight months, according to the company's latest Threat Intelligence report. The attacks targeted by Anthropic include a suspected Russia-linked cyber espionage campaign and efforts by Chinese AI firms to extract and replicate Claude's capabilities.

The use of AI in cyberattacks has expanded beyond simple chatbot interactions, with multi-agent frameworks now executing tasks. Anthropic observed over 151 million exchanges attributed to Alibaba, Moonshot, DeepSeek, and Xiaomi during the period, with peak activity peaking at nearly 3 million per day from over 3,500 accounts believed to be fraudulent.

Chinese tech giant Alibaba ran the largest illicit distillation attack, allegedly aimed at extracting Claude's capabilities to enhance its own Qwen models. Alibaba did not respond to a request for comment. Moonshot and DeepSeek allegedly used live customer conversations containing sensitive information as training data for Claude.

A hacking group modeled after Russia's Midnight Blizzard, linked to the SVR foreign intelligence service, used AI for phishing, hotel Wi-Fi hijacking, and WhatsApp takeover operations against Ukrainian government, military, and diplomatic targets. The group also rewrote malware to evade detection using AI.

Anthropic detected and disrupted activity linked to affiliates of the ShinyHunters cybercrime collective, which has been involved in attacks on major corporations worldwide. The report also identified new categories of threat actors misusing Claude, including those developing software for conventional weapons and targeting intelligence gathering and procurement related to weapons programs in China, Russia, and Yemen.