A memo released last week by Michael Kratsios, director of the White House Office of Science and Technology Policy, alleges that some countries, especially China, have systematically stolen American AI models and systems. The memo, posted by Kratsios on social media, accuses China and other foreign organizations of using thousands of proxies and jailbreaking techniques to copy American AI tools. “AI models built on stolen or unethically obtained information are not reliable,” the post says. “When you steal someone else’s hard-earned invention and build your own technology, you cannot trust the quality and reliability of that technology.” According to CNN, at the heart of it all is a loophole called ‘distillation’, which involves copying years of data and intellectual property accumulated by large language models to produce similar models cheaply and with little technology. Anthropic and OpenAI have also claimed that companies like DeepSec are involved in this loophole and that they have evidence that they have copied the style and technology of major American AI models. The US claims that the data and model outputs of American companies are being extracted and reused to train competing systems, and that this practice is rapidly advancing China’s AI capabilities. However, detailed technical evidence has not been made public with the memo. China has denied such allegations in the past and has claimed that its technological progress is driven by domestic innovation.