Are Chinese companies stealing American AI data? Anthropic appealed to the Trump government

The message sent to diplomatic and consular missions around the world on Friday instructed diplomatic staff to discuss with their foreign counterparts concerns that adversaries are extracting information from US AI models and using it. "distill" Are doing.

Chinese companies, especially AI startup DeepSeek, are making large-scale efforts to steal data from US artificial intelligence labs. "distillation" involves training smaller AI models using the output generated by larger and more expensive AI models. This process is done to reduce the costs associated with training a new and powerful AI tool.

Anthropic complained to the Trump government

The move by the US government comes following a complaint by Anthropic about its actions against three major Chinese AI companies. "Claude" Accused of secretly training competing models by using chatbots on a large scale. Anthropic alleged that Chinese labs DeepSeek, Moonshot AI and MiniMax had tampered with its Claude AI tool in violation of corporate rules.

Zilin Yang, the CEO and founder of Moonshot AI, who has been charged by both Anthropic and the US government. In its public blog post, Anthropic said "We have identified industrial-scale campaigns by three AI labs DeepSeek, Moonshot and MiniMax, which aim to illegally extract Claude’s capabilities to improve their own models. These Labs interacted with Claude approximately 16 million times using approximately 24,000 fake accounts, violating our Terms of Service and regional access restrictions.

serious allegations

The complaint also describes in detail the technology used by Moonshot AI. Using fake accounts and proxy services to access Claude at scale while remaining anonymous. The volume, structure, and focus of their prompts were significantly different from typical usage patterns, pointing to a deliberate attempt to extract capabilities rather than legitimate use.

Moonshot AI alone recorded over 3.4 million interactions. Moonshot’s "Kimi" The model gained access in many different ways, using hundreds of fake accounts. The use of a variety of accounts made it difficult to identify this campaign as a single coordinated action.

What was said in the US State Department cable?

According to an exclusive report by Reuters, a cable from the US State Department states that its purpose is to warn about the risks that are associated with using AI models derived from US-owned AI models and to prepare the basis for possible further actions by the US government.

Chinese AI companies Moonshot AI and MiniMax have been specifically named in this cable. According to the cable, AI models created through secret and unauthorized campaigns help foreigners launch products that are superior to the original systems on some parameters, and do so at much lower costs. They fail to match the full performance capabilities of the original systems.