Anthropic says it disrupted AI abuse campaigns tied to Russia and China
Anthropic says it disrupted malicious uses of its Claude models, including a Russia-linked espionage campaign and distillation attempts by Chinese labs.
Anthropic has disclosed that it disrupted a series of malicious uses of its Claude artificial intelligence models over an eight-month period, including an espionage operation linked to Russia and attempts by Chinese AI developers to replicate the system's abilities.
In its latest threat intelligence report, the company said cybercriminals and state-backed hackers are increasingly turning to AI not merely as an assistant but as the engine that plans and carries out large parts of an attack. In many of the campaigns it observed, humans acted more as supervisors than as hands-on operators.
Anthropic said most of the operations it tracked relied on AI for direct execution or orchestration, with multi-agent frameworks handling tasks rather than simple question-and-answer exchanges with a chatbot.
The company said it disrupted activity from seven China-based laboratories during the period. Among those it named were technology major Alibaba, Moonshot, DeepSeek and Xiaomi.
Operators linked to Alibaba were behind what Anthropic described as the largest "illicit distillation" effort it has seen. The campaign allegedly sought to extract Claude's capabilities and use them to improve Alibaba's own Qwen models. Anthropic said it recorded more than 151 million exchanges tied to Alibaba between May and July 2026, peaking at close to three million a day from over 3,500 accounts it characterised as fraudulent.
Distillation is the practice of training a smaller AI model on the output of a larger, more costly one in order to cut the expense of building a new tool.
Rather than issuing bulk queries, Anthropic alleged that Moonshot, the creator of the Kimi chatbot, and DeepSeek routed live customer conversations — sometimes containing sensitive information — through Claude and used its responses as training data.
Separately, a hacking group whose methods matched those of the Russia-based threat actor Midnight Blizzard allegedly carried out phishing, hotel Wi-Fi hijacking and WhatsApp account takeovers targeting Ukrainian government, military and diplomatic personnel. According to Anthropic, the group used AI at nearly every stage of the operation, including building a system that detected when its malware was flagged by security defences and rewrote the code until it evaded detection again.