White House AI crackdown opens way for Chinese model makers to fill gap

The Trump administration’s crackdown on Anthropic’s leading artificial intelligence model looks like a gift to China, its biggest adversary in the AI ​​race.

After a two-week shutdown due to export control directives, Anthropic received permission from the White House on Friday to release its powerful Mythos 5 model to some companies and federal agencies, but its Fable 5 model remains off the market. Meanwhile, OpenAI also announced on Friday that it would restrict the deployment of GPT 5.6 models in response to government requests.

The two major AI model developers in the US are competing with each other, with the tech giants google By limiting regulatory hurdles, the U.S. government aims to open the door to rapid AI development and develop cutting-edge technology. Doing otherwise would limit domestic AI to serve China’s interests, which are rapidly catching up with the United States, according to many technology executives and Trump administration officials.

But as Anthropic follows the U.S. government’s national security concerns, Chinese companies are launching models that rival Frontier Labs in some features. Zhipu’s GLM 5.2, released earlier this month, can perform on par with top US laboratories on some cyber benchmarks and even rival Mythos’ capabilities, researchers say.

“Many smart people and people in the AI ​​community claim that GLM-5.2 is China’s first AI model that uncompromisingly matches, and often surpasses, public AI models from major US labs,” venture capitalist Marc Andreessen said in a post on X over the weekend. “Incredible timing given current events.”

Sam Bresnick, a researcher at Georgetown’s Center for Security and Emerging Technologies, called the recent developments “a pretty good wake-up call,” and Christopher Wood, a strategist at Jefferies, cited industry sources in a report to clients as saying GLM 5.2 is “roughly on par with Anthropic as an enterprise market competitor, and only a quarter as expensive per token.”

Even former Trump cryptocurrency and AI czar David Sachs, who has been a vocal critic of Anthropic’s approach to AI safety, wrote a cryptic post about X over a screenshot of a Wall Street Journal headline about China rivaling Anthropic in cybersecurity.

“A year ago, President Trump declared that America was in a global AI race, and the way to win it was by driving innovation, driving infrastructure, driving energy, and driving exports,” Sachs wrote. “President Trump was absolutely right, and we are departing from that strategy at our peril.”

Representatives for Anthropic, OpenAI, and the White House did not respond to requests for comment.

The China challenge comes just as most American companies are transitioning from the so-called TokenMax era, where developers can spend unlimited amounts of money on AI, to an era that emphasizes efficiency and return on investment. It also affects China.

Earlier this month, Flo Crivello, CEO of AI startup Lindy, stopped using Anthropic’s Claude model and shifted 100% of her traffic to DeepSeek, a Chinese company that makes a cheaper open-weight alternative.

“We did that, and we saw the cost curve come crashing down into the ground,” Crivello told CNBC in an article published last week.

Chinese AI developers have easy access to US users because it is very easy for companies to download promiscuous weight models and run them on their own servers without relying on a third-party cloud.

“With the zero-gravity model, it’s kind of like the Wild West,” said Travis Lanham, co-founder of AI security startup Armadin. The company is experimenting with the GLM 5.2 and another Chinese model, Moonshoot AI’s Kimi K2.7.

Lanham said these models show improved capabilities for cybersecurity use cases such as analyzing reconnaissance data and writing customer exploit code.

How the two largest economies handle each other’s sensitive hardware has raised new questions in policy circles about whether U.S. authorities will allow this to continue. For years, the U.S. government has done everything possible to keep cutting-edge AI innovation out of China’s hands through U.S. export restrictions on AI chips from China. Nvidia and Advanced Micro Devices. The United States also banned American companies from using Huawei equipment, citing national security concerns.

Last year, the US allowed exports of Nvidia’s H200 chips (the same model used by US companies) to the Chinese region. But Nvidia said earlier this year that it had not yet made any revenue from the chips and that it did not know whether China would allow imports of its products.

Regarding GLM5.2, tesla and space x In a post about X, founder Elon Musk wrote in response to a user’s question about when that milestone would be reached, that the model would probably reach Fable level by the first quarter.

“It won’t take long,” Zhipu founder Jie Tang wrote in a reply to Musk.

Niche brands aren’t the only ones paying attention to Chinese models. like a major company Shopify and airbnb have promoted the benefits of alibabaQwen 3 for expanding AI capabilities. coinbase CEO Brian Armstrong wrote in X last week that the company is leveraging open weight models such as GLM 5.2 and Kimi 2.7, and has been able to cut its AI spending by almost half, despite increasing token usage.

Cybersecurity is a top concern for many industry professionals. Some open weight models are already able to automate many stages of a cyberattack, and Hedo Kobets, CEO of industry startup Silverfort, worries that they are only months away from being fully operational.

“If the U.S. government doesn’t take this opportunity and prepare the industry, no one will be prepared when the Chinese model reaches a similar level,” he said.

—CNBC’s Deirdre Bosa and Ashley Capoot contributed to this report.

WATCH: How Z.ai is closing in on America’s AI frontier