China is flipping the script on the AI fight, accusing American companies of quietly training their systems on Chinese models while Washington moves to punish Beijing for doing the same.
Story Snapshot
- China’s commerce ministry claims many U.S. AI firms “distilled” Chinese models during research and training.
- President Trump’s team accuses Chinese labs of industrial-scale theft of American AI through distillation attacks.
- Cheap, powerful Chinese open models are pulling U.S. companies in, raising serious security and dependency risks.
- New Trump-era crackdowns and possible sanctions aim to protect American innovation and stop foreign exploitation.
China’s New Accusation Against U.S. AI Companies
Chinese officials are now accusing American artificial intelligence companies of doing the very thing Washington says China has done for years. Beijing’s Ministry of Commerce said it is “understood that many American artificial intelligence enterprises have distilled Chinese models during their research, development and training processes,” claiming U.S. firms used Chinese examples to train their systems without proper acknowledgment. The ministry blasted U.S. charges against Chinese labs as lacking factual basis and called them “typical acts of artificial intelligence hegemonism.”
Distillation is a common method in AI, where a weaker “student” model learns from a stronger “teacher” by copying its outputs. Chinese officials argue that if Washington says Chinese use of distillation is theft, then American use of Chinese models should be judged by the same standard. Their statement came just days after reports that the United States is weighing new curbs or even a ban on foreign open models built through distillation, especially those tied to Chinese firms. Beijing is clearly trying to paint U.S. policy as hypocritical and politically driven.
Trump Administration’s Crackdown on Chinese AI Theft
President Trump’s administration has not been shy about calling out Chinese AI firms for stealing American technology. A memo from the White House Office of Science and Technology Policy said foreign entities “predominantly located in China” were running “industrial-scale campaigns” to copy cutting-edge U.S. models. These operations allegedly use tens of thousands of fake accounts and special tools to extract model outputs and turn them into rival systems, all without permission from the American companies that built the originals.
Leading U.S. labs like OpenAI and Anthropic have backed up those claims with warnings to lawmakers. OpenAI told Congress that Chinese startup DeepSeek used unauthorized distillation against its systems, and Anthropic said DeepSeek, Moonshot, and MiniMax ran mass campaigns to harvest outputs from its Claude model. Reports describe tens of thousands of accounts and millions of conversations being used to “illicitly extract” capabilities, then repackage them into Chinese models sold at a steep discount. That is not normal competition; it looks like coordinated technology theft that costs American companies billions in lost value.
Cheap Chinese Models and U.S. Corporate Temptation
While Washington sounds the alarm, many U.S. businesses are quietly turning to Chinese-built AI to save money. Chinese open models are often “60% to 90% cheaper” than top American systems from OpenAI and Anthropic, according to one U.S. data expert. Policy analysts estimate they are only six to nine months behind America’s frontier models in performance, yet cost a fraction as much to run. For cost-cutting executives, that bargain can be hard to resist, even when it raises national-security and privacy concerns.
That rush toward low-cost Chinese tools is exactly what worries Trump officials and many conservative lawmakers. If Chinese labs gained their edge by scraping American models through distillation attacks, then every U.S. company that deploys those tools may be rewarding theft. It also risks putting sensitive customer data, corporate secrets, and even government-related information into systems built and controlled under Beijing’s rules. For a country that has seen China steal defense and tech secrets for decades, letting U.S. firms rely on those tools could be a dangerous form of digital dependence.
Double Standards, Open Models, and the Fight for Fair Rules
China’s commerce ministry is trying to turn this into a narrative of American double standards. Officials argue that U.S. companies also distill Chinese models, and that some Chinese systems are already “in a leading position,” so Washington’s theft claims must be political. But industry reporting notes that many Chinese models are open source or “open weight,” meaning anyone can download or query them under broad licenses. Distilling an openly licensed model is very different from secretly scraping a proprietary system that bans copying in its terms of service.
Daily AI Briefing Report (US/EU/China) – July 28, 2026
1/ Top 5 AI News of the Day
• Nvidia launches Open Secure AI Alliance with 40+ firms for open security tools
• Moonshot releases Kimi K3 weights (2.8T params, largest open-weight model)
• Nvidia invests billions in…— El Mundi (@El_Mundi_X) July 28, 2026
That distinction matters for conservatives who believe in both property rights and open competition. President Trump’s team has stressed that distillation itself is a normal tool when used with consent and proper licensing, but becomes abuse when done “illicitly” and “at industrial scale” against protected frontier models. The real question is whether Beijing will ever allow transparent, independent checks on how its top labs trained their systems. Until Chinese firms and officials open their books, U.S. leaders are right to guard American innovation, push for sanctions on cheaters, and insist that any use of foreign AI in this country respects our rules, our security, and our sovereignty.
Sources:
insiderpaper.com, reuters.com, bbc.com, nextgov.com, cnbc.com, nbcnews.com, globaltimes.cn, docs.house.gov, nypost.com
