The Trump administration is signaling a major escalation in the U.S.-China artificial intelligence rivalry, shifting its focus from restricting advanced chips to potentially targeting Chinese AI models themselves if they are found to have been built using stolen American intellectual property.

U.S. Treasury Secretary Scott Bessent said on Tuesday that Washington would examine Chinese open-source AI models for signs of intellectual property theft, warning that the administration could impose sanctions on Chinese AI companies if such practices are established.

“We’ve seen a lot of talk about open source models coming and threatening the large language models in the US,” Bessent said during an interview on Fox Business. “This administration supports open source models, but what we do not support is IP theft. If we see, especially, that overseas models are stealing from our great companies, we have the ability to sanction them because of this theft.”

His remarks, first reported by Bloomberg, come at a pivotal moment in the global AI race, with Chinese developers rapidly closing the gap with U.S. frontier AI companies. The comments also suggest the White House is broadening its strategy for maintaining America’s technological lead beyond export controls on semiconductors to the AI software layer itself.

Since 2022, Washington has relied heavily on export restrictions aimed at limiting China’s access to Nvidia’s most advanced AI chips and semiconductor manufacturing equipment. Those measures were designed to slow China’s ability to train cutting-edge AI models.

However, Chinese companies have responded by optimizing software, developing more efficient AI architectures and building increasingly capable models using domestically available hardware.

The latest wave of Chinese models has surprised many industry observers.

Moonshot AI’s Kimi K3, unveiled last week, has been described by the company as the world’s largest open-weight AI model with 2.8 trillion parameters. Other firms including Alibaba, Z.ai, MiniMax and DeepSeek have also introduced competitive open-weight systems this year, many of which deliver performance approaching leading U.S. proprietary models while costing significantly less to deploy.

Their rapid progress has challenged a long-standing assumption among Western analysts that Chinese AI firms lagged their U.S. counterparts by six months or more. As Chinese models gain adoption among enterprises seeking lower-cost alternatives, pressure has mounted on companies such as OpenAI and Anthropic, whose business models depend on premium-priced proprietary AI services.

At the center of Washington’s concerns is model distillation, a technique that enables developers to transfer some of the capabilities of a larger AI model into a smaller and cheaper system. Major U.S. AI companies have repeatedly lamented that foreign competitors are using distillation to replicate the behavior of their frontier models without bearing the enormous costs of developing them from scratch.

The White House said in April it would work with leading AI developers to combat theft of AI technology, acknowledging growing concern that America’s competitive advantage could erode if its models are copied overseas.

Earlier this week, Axios reported that the Trump administration is considering a broad prohibition on Chinese open-source AI models, although others familiar with policy discussions have disputed that such a sweeping ban is imminent.

If sanctions are ultimately imposed, they could represent the first time Washington has directly targeted AI models rather than the hardware used to train them, opening a new front in the technology rivalry between the world’s two largest economies.

Industry Divided Over What Constitutes Theft

Whether model distillation amounts to intellectual property theft remains one of the most contentious issues in artificial intelligence. Many AI companies argue that copying the behavior of proprietary models without authorization undermines years of expensive research and development.

Others disagree.

Microsoft Chief Executive Satya Nadella recently questioned the industry’s position, arguing that AI companies themselves rely on broad access to publicly available information when training their own systems.

“While the great innovation that comes from model providers having fair use rights to train models on public data is needed, I find it ironic that the status quo is to then turn around and impose restrictive terms on distillation,” Nadella said earlier this month.

Some researchers also argue that distillation alone cannot explain China’s recent advances.

Hugging Face Chief Executive Clem Delangue said China’s success stems primarily from strong domestic research capabilities rather than copying Western systems.

“We know distillation to be a very small factor in the ability to create good models, and it’s a practice that everyone is doing, including companies in the U.S.,” Delangue said on TechCrunch’s Equity podcast.

“If it were easy just to do distillation to get good at building AI models, there would be many other countries, including in the U.S., with much better open source AI. The reality is they have really, really good research teams in China… taking a much more open and collaborative approach to AI than in the U.S.”

However, Bessent’s comments indicate that Washington increasingly sees advanced AI models as strategic assets comparable to advanced semiconductors. That position mirrors Beijing’s own evolving policy. Chinese authorities are reportedly considering tighter export controls on advanced AI technologies and model weights while exploring measures to limit overseas access to China’s most advanced AI systems.

As both governments move to protect their domestic AI champions, the competition is evolving beyond hardware and into control over the software, algorithms and data that underpin next-generation artificial intelligence.

If the United States ultimately sanctions Chinese AI developers over intellectual property concerns, it would mark a significant expansion of the technology conflict, potentially affecting the global adoption of Chinese open-source models and further fragmenting the international AI ecosystem.

A U.S. federal judge has granted final approval to Anthropic’s landmark $1.5 billion settlement with a class of authors who accused the artificial intelligence company of unlawfully using their books to train its Claude chatbot, bringing to a close one of the most closely watched copyright disputes in the AI industry and setting an important benchmark for dozens of similar lawsuits against major technology companies.

U.S. District Judge Araceli Martinez-Olguin in San Francisco on Monday approved the settlement, rejecting objections from authors who argued that the payout was insufficient. The agreement is the largest known settlement in a U.S. copyright case and the first major AI copyright lawsuit involving generative AI training to reach a negotiated resolution.

The case has been widely viewed as a bellwether for the legal battles unfolding between copyright holders and AI developers over the use of books, news articles, music and other creative works to train large language models.

Judge William Alsup, who presided over much of the litigation before retiring, had granted preliminary approval to the settlement last September.

“We reached this settlement in 2025, after the court’s landmark ruling that training AI on books is fair use under copyright law, which remains the law today,” Anthropic Deputy General Counsel Aparna Sridhar said in a statement.

“We are pleased that more than 91% of authors and publishers covered by the settlement have claimed their share of the payment, and we’re looking forward to bringing this matter to a close.”

Lead plaintiffs’ attorney Justin Nelson described the agreement as a milestone for copyright enforcement.

“It is the largest known copyright recovery in history. We look forward to making distributions to the Class as promptly as possible,” Nelson said.

The litigation began in 2024, when a group of authors sued Anthropic, alleging the company copied pirated versions of their books without authorization to train Claude, its flagship generative AI model.

Anthropic, which is backed by Amazon and Alphabet, argued that using copyrighted books for AI training constituted fair use, a long-established doctrine in U.S. copyright law permitting limited use of protected works under certain circumstances.

In a landmark ruling last June, Judge Alsup largely agreed with Anthropic’s position, concluding that training AI models on copyrighted books was a transformative use protected under the fair use doctrine. The decision represented one of the most significant judicial victories for AI developers and has become a key legal precedent as courts consider similar claims against companies including OpenAI, Meta, Microsoft, Google and others.

However, Alsup also found that Anthropic infringed copyright by maintaining a digital repository containing more than 7 million pirated books, describing the company’s “central library” as distinct from the AI training process itself because many of the works were retained without necessarily being used to train Claude.

That ruling left Anthropic exposed to potentially enormous statutory damages.

Before the settlement was reached, the case was scheduled to proceed to trial last December to determine damages related to the storage of the pirated books. Because U.S. copyright law allows statutory damages of up to $150,000 per infringed work in cases involving willful infringement, legal analysts estimated Anthropic could theoretically have faced liabilities running into the hundreds of billions of dollars, although actual awards in copyright litigation are typically far lower.

The settlement eliminates that uncertainty while allowing Anthropic to avoid years of additional litigation and potential appeals. The agreement also provides significant compensation to participating authors and publishers without requiring them to prove individual damages.

According to Anthropic, more than 91% of eligible copyright holders have already claimed their share of the settlement fund, reflecting broad participation despite objections from a minority of authors.

Several authors challenged the settlement, arguing that the compensation failed to reflect the scale of Anthropic’s alleged infringement. Others contended that the agreement unfairly excluded certain copyright owners or awarded excessive legal fees to the plaintiffs’ attorneys.

Judge Martinez-Olguin rejected those objections, finding that the settlement represented a reasonable outcome given the litigation risks facing both sides.

The judge wrote that criticisms regarding the settlement amount were “not grounded in a realistic assessment of the overall risks and rewards of a trial.”

She also approved more than $101 million in attorneys’ fees, substantially below the $187.5 million requested by class counsel.

The settlement comes as AI developers face mounting legal challenges over the datasets used to train sophisticated generative AI systems. Publishers, authors, musicians, artists and media organizations have argued that technology companies have built commercially valuable AI products using copyrighted material without obtaining licenses or providing compensation.

Technology companies counter that AI training is fundamentally transformative, does not reproduce the original works for consumers, and therefore qualifies as fair use.

The Anthropic case is particularly significant because it produced one of the first major judicial rulings recognizing AI model training as fair use while simultaneously finding liability for maintaining unauthorized copies of copyrighted works. That distinction is likely to influence ongoing litigation across the United States as courts seek to balance copyright protections with technological innovation.

The settlement does not resolve all of Anthropic’s copyright disputes. Some authors and publishers opted out of the class action and are continuing to pursue separate lawsuits against the company, meaning additional legal battles over AI training practices remain underway.