OpenAI and Anthropic say Chinese labs copied US AI models
OpenAI and Anthropic have called out Chinese labs like DeepSeek for using "distillation attacks": basically, tricks to copy American AI models by sending tons of prompts and learning from the responses.
Anthropic says some Chinese companies made 24,000 fake accounts and hit their chatbots over 16 million times, which breaks the rules.
Chris McGuire from the Council on Foreign Relations summed it up as stealing "the weights of our best AI models," which he called some of the most valuable tech out there.
US firms seek government action
Chinese labs have been using these shortcuts to quickly catch up with US tech, even training new models on NVIDIA chips in violation of export controls.
As the gap narrows, American companies worry about losing their edge—and are now pushing for government action to protect their huge investments in AI innovation.