Anthropic Blocks Foreign Cyber Attacks and AI Theft

SAN FRANCISCO — Artificial intelligence firm Anthropic announced that it has successfully disrupted a series of sophisticated malicious campaigns targeting its Claude AI models over the past eight months. According to the company's latest Threat Intelligence report, the thwarted operations included a suspected Russia-linked cyber espionage campaign alongside extensive efforts by Chinese artificial intelligence firms attempting to illegally extract and replicate Claude's advanced capabilities.

The findings highlight a growing trend of state-backed actors and cybercriminals leveraging artificial intelligence not merely as a conversational assistant, but to autonomously orchestrate and execute complex cyberattacks. Anthropic revealed that operators in these campaigns frequently utilized multi-agent frameworks, positioning humans as mere overseers rather than hands-on supervisors. 

Among the major disruptions, the company pointed to a hacking collective whose tradecraft aligned with the Russia-based threat actor Midnight Blizzard. This group allegedly deployed AI across nearly every stage of phishing, hotel Wi-Fi hijacking, and WhatsApp-takeover operations targeting Ukrainian government, military, and diplomatic personnel. The actors reportedly used AI to automatically detect when their malware was flagged by security defenses and instantly rewrite the code to evade detection.

Concurrently, Anthropic detailed expansive intellectual property maneuvers originating from China, accusing labs and tech giants of engaging in massive "illicit distillation" attacks. These unauthorized operations were allegedly designed to harvest outputs from Claude models to train and upgrade domestic models, such as Alibaba's Qwen series. Anthropic reported logging over 151 million suspicious exchanges tied to Alibaba across thousands of fraudulent accounts between May and July. 

Additional firms, including Moonshot and DeepSeek, were accused of routing live customer conversations through Claude to siphon training data. Security experts warn that these multi-front threats underscore the escalating race to safeguard foundational AI infrastructure against weaponization and intellectual theft by foreign adversaries.