US Agencies Accuse Chinese AI Firms of Distillation
US security and intelligence agencies announced that Chinese artificial intelligence companies are copying capabilities from American AI models.
United States cybersecurity and intelligence agencies have announced that China-based artificial intelligence companies are conducting industrial-scale distillation attacks and systematically extracting proprietary features of American AI models.
Details of the Joint Security Warning
In a joint bulletin published by the National Security Agency, the Cybersecurity and Infrastructure Security Agency, and the Federal Bureau of Investigation, it was emphasized that the activities of Chinese firms form the core of their artificial intelligence development strategies.
Targeted American Models
The joint report stated that Chinese AI firms such as DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI have extracted billions of tokens since late 2024.
It was noted that various leading American models, including Anthropic Claude, OpenAI GPT, Google Gemini, and SpaceXAI Grok, were targeted in this scope.
Background of the Operations
It was reported that these operations are likely conducted with the support of the Chinese government and violate terms of use.
It was recorded that the companies use bulk subscriptions, chain-of-thought reasoning extraction methods, and black-market proxies to bypass geographical restrictions and detection mechanisms.