US and AI firms say China-based labs are copying chatbots at scale

In short: U.S. officials and major AI companies say China-based groups are using a long-known method to copy how American chatbots respond, by asking them huge numbers of questions and training imitators on the answers.

What's going on

U.S. officials, including the White House Office of Science and Technology Policy, say foreign actors mostly based in China are running large campaigns to copy the abilities of top U.S. AI models. The claim is that these groups use public access points like APIs, which are basically doorways that let software talk to a service.

Officials say the groups create tens of thousands of fake or proxy accounts and then send millions of automated prompts to systems from companies like OpenAI and Anthropic. They collect the responses and use them as training material to build a new model that behaves similarly.

This method is often called “knowledge distillation” or “model extraction.” In normal, approved use, distillation is like making a smaller, faster apprentice learn by watching a bigger expert. The dispute here is about doing it without permission, using massive automated copying, and then turning the results into competing products.

OpenAI has told U.S. lawmakers that China-based firm DeepSeek was trying to replicate the performance of its models through distillation. Anthropic says it has seen large-scale attempts from DeepSeek, Moonshot AI, and MiniMax, and it describes activity involving tens of thousands of accounts and tens of millions of conversations.

What to watch

This sits in a legal grey area, because distillation itself is not new or automatically illegal, and outcomes may depend on contract terms and computer misuse rules. U.S. officials are considering tighter access controls and other steps, including possible sanctions. Chinese state-linked media has pushed back, calling distillation a normal part of competition.

Source: NYTimes