Trump Administration Plans Crackdown On Chinese Use Of U.S. AI Models
The Trump administration on Thursday in Washington announced a crackdown on Chinese companies it accuses of exploiting U.S. artificial intelligence models, ordering steps to identify, block and punish large-scale model extraction.
A memo from White House aide Michael Kratsios accused foreign entities "principally based in China" of industrial-scale campaigns to distill capabilities from U.S. frontier models. The administration said it will work directly with American AI firms to spot such activity, build technical defenses and craft punishment mechanisms including possible sanctions. Anthropic flagged the scale of the problem, reporting more than 24,000 fraudulent accounts that generated over 16 million exchanges to extract capabilities from its Claude model. China's embassy in Washington and the Foreign Ministry in Beijing called the accusations groundless and said the move unfairly targets Chinese firms. The House Foreign Affairs Committee advanced a bipartisan bill to identify actors extracting "key technical features" of closed-source models and to authorize penalties; Rep. Bill Huizenga called such attacks a new frontier of Chinese economic coercion and intellectual property theft.
The episode traces back to U.S. export controls announced in October 2022 that restricted high-end chips and GPUs bound for China. Those curbs expanded in 2023 as officials worried about China gaining military and AI advantages through better hardware. Critics warned software and cloud access could evade hardware rules, a fear that grew after a January 2025 claim that startup DeepSeek distilled outputs from OpenAI models. By April 2026 a Stanford report found the performance gap between top U.S. and Chinese models had effectively closed, increasing pressure to extend controls beyond silicon. Meanwhile the U.S. still controls roughly 74 percent of global high-end AI compute capacity, while China holds about 14 percent, and Chinese models had lagged by about seven months on average since 2023 until recent narrowing.
Early U.S. steps targeted hardware, but recent reporting shifted the debate toward software, cloud services and model theft. Coverage from national outlets and the Stanford paper pushed lawmakers to consider new legal tools and penalties for extraction attacks. Some social media voices see the move as pressure ahead of a planned presidential trip to China, while others call it an overreach that limits foreign AI choices.
The administration says it will rely on private-sector cooperation to build defenses and carry out punishments, though exact enforcement steps and legal authority remain unclear. China rejected the allegations as unjustified suppression, and the dispute risks accelerating an AI arms race between the two countries.
📊 Relevant Data
Chinese AI companies DeepSeek, Moonshot AI, and MiniMax conducted large-scale knowledge distillation attacks on U.S. frontier models, with Anthropic detecting over 24,000 fraudulent accounts used to generate more than 16 million exchanges to extract capabilities from its Claude model.
Detecting and Preventing Distillation Attacks — Anthropic
As of October 2025, the U.S. controls approximately 74 percent of global high-end AI compute capacity, compared to China’s 14 percent.
The State of AI Competition in Advanced Economies — Federal Reserve
Chinese AI models have lagged behind U.S. frontier models by approximately seven months on average since 2023.
US vs China ECI — Epoch AI
📌 Key Facts
- In a Thursday memo, Michael Kratsios accused foreign entities "principally based in China" of industrial-scale campaigns to distill capabilities from leading U.S. AI models, and the administration said it will work directly with American AI companies to identify such activities, build technical defenses and develop punishment mechanisms.
- The House Foreign Affairs Committee unanimously advanced a bipartisan bill to create a formal process to identify foreign actors extracting "key technical features" of closed-source U.S. AI models and to punish them, including via sanctions.
- Rep. Bill Huizenga described "model extraction attacks" as a new frontier of Chinese economic coercion and theft of U.S. intellectual property.
- China’s embassy in Washington and the Foreign Ministry in Beijing publicly rejected the U.S. allegations as "groundless" and an "unjustified suppression" of Chinese companies.
- Context cited in the reporting: a Stanford Institute for Human-Centered AI report found the U.S.-China performance gap for top AI models has "effectively closed," and the article recalls that Chinese startup DeepSeek rattled U.S. markets with a low-cost large language model amid prior allegations (from then–White House adviser David Sacks) that it distilled OpenAI models.
📊 Analysis & Commentary (1)
"A Garicano-style deep-dive arguing that while U.S. concerns about foreign model extraction justify targeted safeguards, the correct economic response combines narrowly tailored protections with big public investments in skills, competition policy and international standards rather than broad export bans."
📰 Source Timeline (2)
Follow how coverage of this story developed over time
- Michael Kratsios issued a Thursday memo accusing foreign entities 'principally based in China' of industrial-scale campaigns to distill capabilities from leading U.S. AI models.
- The administration says it will work directly with American AI companies to identify such activities, build technical defenses, and develop punishment mechanisms.
- China's embassy in Washington and the Foreign Ministry in Beijing publicly rejected the U.S. allegations as 'groundless' and 'unjustified suppression' of Chinese companies.
- The House Foreign Affairs Committee unanimously advanced a bipartisan bill to create a formal process to identify foreign actors extracting 'key technical features' of closed-source U.S. AI models and punish them, including via sanctions.
- Rep. Bill Huizenga described 'model extraction attacks' as a new frontier of Chinese economic coercion and theft of U.S. intellectual property.
- The article cites a Stanford Institute for Human-Centered AI report finding that the U.S.-China performance gap for top AI models has 'effectively closed.'
- The piece recalls that Chinese startup DeepSeek rattled U.S. markets with a low-cost large language model and notes prior allegations from then–White House adviser David Sacks that it distilled OpenAI models.