China and US Face AI Showdown Over Model Distillation Accusations and Investment Restrictions

Gate News message, April 24 — China has rejected U.S. accusations that its tech giants are exploiting American AI technology through industrial-scale distillation, as both countries set up for a major collision over AI development and investment control. The Trump administration is preparing to crack down on entities accused of stealing American AI models, while Beijing is reportedly planning to restrict domestic tech companies from accepting U.S. investment without government approval.

According to Michael Kratsios, Director of the White House Office of Science and Technology Policy, entities "principally based in China" are running deliberate, industrial-scale campaigns to distill U.S. frontier AI systems. Distillation is a method where developers use outputs from larger, more powerful AI models to train smaller, cheaper ones; the U.S. claims this constitutes theft. The White House outlined four specific measures: sharing intelligence on distillation tactics with U.S. AI companies, coordinating defenses, developing best practices for identifying attacks, and exploring accountability mechanisms. OpenAI and Anthropic have previously alleged that Chinese labs used distillation to replicate their models, with Anthropic specifically accusing DeepSeek, Moonshot AI, and MiniMax. Beijing fired back, calling the accusations an "unjustified suppression" of its companies.

China's investment restrictions stem from Meta's approximately $2 billion acquisition of AI startup Manus, which Chinese authorities viewed as a strategic asset loss to a geopolitical rival. Regulators, including the National Development and Reform Commission, are now reportedly telling AI startups like Moonshot AI and StepFun to reject U.S. capital unless Beijing explicitly approves it.

Meanwhile, DeepSeek released a preview of its new V4 model on Friday, claiming it "significantly leads other open-source models" in world knowledge benchmarks and is "only slightly outperformed" by Google's Gemini-Pro-3.1. The V4 is priced at 2 yuan ($0.28) per 1 million tokens, over 100 times cheaper than GPT-5.5 at 216 yuan ($30). A recent Stanford University report noted that the performance gap between top U.S. and Chinese AI models has "effectively closed."