AI firm Anthropic has now publicly accused three prominent Chinese AI companies—DeepSeek, MiniMax, and Moonshot—of running large-scale, coordinated campaigns to “illicitly extract” capabilities from its Claude models through a technique known as distillation. In a detailed blog post, the company said the three labs used more than 24,000 fraudulent accounts to generate over 16 million exchanges with Claude, its flagship AI chatbot.
The accusations add to mounting US concerns about Chinese labs rapidly closing the capability gap with American frontier models without incurring the full cost of independent research and development. OpenAI raised similar alarms in a memo to the House Select Committee on China earlier this month, accusing DeepSeek and others of using “obfuscated” distillation techniques to “free-ride” on OpenAI’s investments.
“We have identified industrial-scale campaigns by three AI laboratories—DeepSeek, Moonshot, and MiniMax—to illicitly extract Claude’s capabilities to improve their own models. These labs generated over 16 million exchanges with Claude through approximately 24,000 fraudulent accounts, in violation of our terms of service and regional access restrictions. These labs used a technique called “distillation,” which involves training a less capable model on the outputs of a stronger one. Distillation is a widely used and legitimate training method. For example, frontier AI labs routinely distill their own models to create smaller, cheaper versions for their customers. But distillation can also be used for illicit purposes: competitors can use it to acquire powerful capabilities from other labs in a fraction of the time, and at a fraction of the cost, that it would take to develop them independently,” the blog post read.
To provide some context, distillation involves training a smaller or more efficient model on the outputs of a larger, more capable one. While the method is a standard and legitimate technique for model compression, Anthropic says the Chinese firms violated its terms of service by using fraudulent accounts, proxy networks, and other obfuscation methods to evade detection and rate limits while systematically harvesting Claude’s responses at an industrial scale. The company claims it identified the campaigns with “high confidence” using IP address patterns, metadata analysis, behavioral signals, and corroboration from unnamed industry partners who observed similar activity on their own platforms.
DeepSeek is accused of having generated more than 150,000 exchanges, focusing on foundational reasoning, alignment, and generating “censorship-safe alternatives” to politically sensitive queries. MiniMax is accused of having produced more than 13 million exchanges, heavily targeting agentic reasoning, tool use, coding, data analysis, computer-use agent development, and computer vision capabilities.
MoonShot, according to the AI firm, accounts for more than 3.4 million exchanges, with emphasis on agentic coding, tool orchestration, and multi-step reasoning. Anthropic revealed that these efforts were not isolated incidents but part of “industrial-scale campaigns” that are “growing in intensity and sophistication.” MiniMax went public in January, while Moonshot is reportedly targeting a $10 billion valuation in its current funding round. All three have benefited from China’s aggressive domestic compute build-out and access to open-source weights from models like Llama and Qwen.
The Tech Portal is published by Blue Box Media Private Limited. Our investors have no influence over our reporting. Read our full Ownership and Funding Disclosure →
