The U.S. government has accused six Chinese artificial intelligence companies of carrying out large-scale campaigns to extract capabilities from advanced American AI models. The companies named are DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI. According to the NSA, FBI and CISA, these firms extracted billions of tokens through millions of exchanges and requests from U.S. frontier AI models. The activity has reportedly been taking place since at least late 2024 and was likely carried out with awareness from the Chinese government.
The agencies described the activity as industrial-scale knowledge distillation, a technique in which one AI system learns from the outputs of another, more advanced system. Distillation itself is a legitimate method used across the AI industry, but U.S. officials say these companies took the technique much further. Their advisory says the campaigns were designed to extract restricted capabilities and proprietary functions from leading American AI systems. The targeted models included versions of Claude, GPT, Gemini and Grok.
U.S. officials say the Chinese companies used millions of carefully designed requests to collect information about coding, mathematics, reasoning and other specialized abilities. The advisory says the companies also used different accounts, proxy services and multiple access routes to continue their operations when restrictions were introduced. Some reportedly used bulk purchases of premium AI subscriptions to reduce costs while allowing large numbers of developers to access American models. These methods, according to the agencies, helped the companies gather large amounts of model output at a much lower cost than developing equivalent systems from scratch.
The advisory provides company-specific allegations as well. It says DeepSeek conducted organized campaigns targeting reasoning capabilities and other specialized functions, while Alibaba used large-scale distillation to improve its Qwen family of models. Moonshot AI is accused of using data from American models to improve its Kimi systems, while MiniMax, StepFun and Z.AI were also linked to similar campaigns. The agencies say the activity targeted areas including software engineering, mathematical reasoning, supervised fine-tuning, reinforcement learning and agentic capabilities.
According to the U.S. agencies, these operations could shorten the time and reduce the money required for Chinese companies to develop advanced AI systems. Instead of building every capability independently, companies can use large quantities of responses from stronger models as training material for their own systems. The U.S. government says this creates economic and strategic risks because proprietary capabilities developed through significant research and investment can be reproduced or transferred without the same level of investment. The agencies also warned that the activity could weaken America’s advantage in frontier AI development.
China has rejected the U.S. accusations, calling them groundless and arguing that distillation is a normal and widely used AI development technique. Chinese officials have accused Washington of trying to restrict competition and maintain control over advanced AI technology. The dispute comes as tensions between the two countries over artificial intelligence continue to grow, with AI expected to remain an important issue in upcoming discussions between U.S. President Donald Trump and Chinese President Xi Jinping. For now, the allegations remain claims made by U.S. agencies, while the Chinese companies and government reject the accusations.
Stay alert, and keep your security measures updated!
Source: Follow cybersecurity88 on X and LinkedIn for the latest cybersecurity news
