Dario Amodei, co-founder and chief govt officer of Anthropic, on the AI Affect Summit in New Delhi, India, on Thursday, Feb. 19, 2026.
Prakash Singh | Bloomberg | Getty Photographs
The Trump administration’s crackdown on Anthropic’s main synthetic intelligence fashions is wanting like a present to the nation’s high adversary within the AI race: China
After a two-week shutdown on account of an export management directive, Anthropic was allowed by the White Home on Friday to release its powerful Mythos 5 model to some corporations and federal companies, although its Fable 5 mannequin stays off the market. OpenAI, in the meantime, additionally mentioned on Friday it could restrict the rollout of its GPT 5.6 models following a authorities request.
The 2 main U.S. AI mannequin builders have been in a race towards one another and tech giants like Google to develop probably the most superior know-how, with the U.S. authorities opening the door to speedy AI growth by limiting regulatory hurdles. Doing in any other case, in keeping with many tech execs and Trump administration officers, would limit home AI to the advantage of China, which has been quickly catching as much as the U.S.
However as Anthropic adheres to the U.S. authorities’s nationwide safety considerations, Chinese language corporations are launching fashions that rival frontier labs in some capabilities. In accordance with researchers, Zhipu’s GLM 5.2, launched earlier this month, can perform on par with high U.S. labs on some cyber benchmarks, even equaling Mythos’ capabilities.
“Many sensible individuals/AI insiders are saying GLM-5.2 is the primary Chinese language AI mannequin to match and infrequently beat the American huge lab public AI fashions with no compromises,” wrote enterprise capitalist Marc Andreessen, in a publish on X over the weekend. “Unbelievable timing given present occasions.”
Sam Bresnick, a analysis fellow at Georgetown’s Middle for Safety and Rising Expertise, known as the current developments “a fairly good wake-up name,” and Jefferies strategist Christopher Wooden wrote in a report to purchasers, citing business sources, that GLM 5.2 “is sort of equal to Anthropic as a competitor for the company market and is only one quarter of the fee by way of value per token.”
Even former Trump crypto and AI czar David Sacks, who’s been a vocal critic of Anthropic’s method to AI security, wrote a cryptic publish on X above a screenshot of a Wall Road Journal headline stating that China has matched Anthropic in cybersecurity.
“A yr in the past, President Trump declared that America was in a world AI race and that the way in which to win it was to be pro-innovation, pro-infrastructure, pro-energy, and pro-export,” Sacks wrote. “President Trump was precisely proper; we deviate from that technique at our peril.”
Representatives from Anthropic, OpenAI and the White Home did not reply to requests for remark.
The Chinese language conundrum lands simply as a lot of company America is shifting from an period of so-called tokenmaxxing, or permitting builders to spend on AI with out restraint, to a deal with effectivity and return on funding. That additionally performs into China’s arms.
Earlier this month, Flo Crivello, CEO of AI startup Lindy, switched his firm off Anthropic’s Claude fashions and moved 100% of its visitors to DeepSeek, a Chinese language firm that makes cheaper, open-weight options.
“We did it, and you would see that value curve go down, like, crash to the bottom,” Crivello informed CNBC for a story revealed final week.
‘Wild West’
Chinese language AI builders are reaching U.S. customers with ease due to how easy it’s for an organization to obtain open-weight fashions and run them on their very own servers with out counting on a third-party cloud.
“With the open-weight fashions, it is form of the Wild West,” mentioned Travis Lanham, co-founder of AI safety startup Armadin, which is experimenting with GLM 5.2 and one other Chinese language mannequin, Kimi K2.7 from Moonshoot AI.
Lanham mentioned the fashions are displaying improved capabilities for cybersecurity use instances like analyzing reconnaissance information and creating buyer exploit code.
Whether or not U.S. authorities will permit that to proceed is an rising query in coverage circles on account of how the 2 largest economies deal with one another’s delicate {hardware}. For years, the U.S. authorities has gone to nice lengths to maintain cutting-edge AI innovation out of China’s arms via U.S. export controls on AI chips from Nvidia and Superior Micro Units. The U.S. additionally banned American corporations from utilizing Huawei tools due to nationwide safety fears.
Final yr, the U.S. cleared Nvidia’s H200 chip — the identical mannequin utilized by U.S. corporations — for export to the China area. However Nvidia said earlier this yr that it had but to generate any income from the chips and that it did not know whether or not China would permit imports of its merchandise.
As for GLM 5.2, Tesla and SpaceX founder Elon Musk wrote in a publish on X that the mannequin would in all probability attain Fable ranges by the primary quarter, responding to a query from a consumer asking when that milestone could be reached.
Zhipu founder Jie Tang, in a reply to Musk, wrote, “will not take that lengthy.”
It is not simply area of interest manufacturers turning to Chinese language fashions. Main corporations like Shopify and Airbnb have touted the advantages of Alibaba‘s Qwen 3 for scaling AI options. Coinbase CEO Brian Armstrong wrote on X final week that his firm is using open-weight fashions resembling GLM 5.2 and Kimi 2.7, permitting it to chop almost half its AI spending regardless of better token use.
Cybersecurity is the first concern for a lot of business specialists. Some open-weight fashions can already automate many phases of a cyberattack, and Hed Kovetz, CEO of business startup Silverfort, worries they’re solely months away from working a whole operation.
“If the U.S. authorities doesn’t let the business make the most of this chance to prepare, then when the Chinese language fashions attain an analogous stage, nobody shall be ready,” he mentioned.
— CNBC’s Deirdre Bosa and Ashley Capoot contributed to this report.
