Nearly half the AI models on a Hong Kong platform that accepts the Trump family's stablecoin were built by Chinese companies the Trump administration has flagged as national security risks.
A Reuters investigation published Monday, August 17, by Lawrence Delevingne counted 43 of the 90 models on WorldClaw's website as products of Alibaba, Baidu, Z.ai and other Chinese firms the administration says pose risks to national security and intellectual property. WorldClaw, founded earlier this year, also offers dozens of models from U.S. labs including OpenAI and Anthropic.
Reuters was explicit on the legal question: "There's nothing illegal about World Liberty's collaboration with WorldClaw or with WorldClaw's relationships with the Chinese companies." But seven experts on Chinese technology, trade and government ethics told Reuters the arrangement sits awkwardly against the administration's own posture.
"As the U.S. government tries to respond to the rise and threat of Chinese AI, it seems hypocritical to go out through WorldClaw to use these tools from China to try and make a bunch of money," said Sam Bresnick, a fellow at Georgetown University's Center for Security and Emerging Technology.
What The Arrangement Actually Is
The connective tissue is a payment rail. WorldClaw accepts USD1, World Liberty Financial's dollar-pegged stablecoin, as one of several crypto options. USD1 is backed by assets such as U.S. Treasuries, and per Reuters the Trumps — who own 38% of World Liberty — are entitled to a percentage of the interest earned on those reserves.
Reuters could not establish the details of that arrangement, or how much the family has earned from platform payments; earnings from token sales exceed $1.4 billion, it reported in June. The president's two eldest sons co-founded the firm; he retains ownership through a trust.
Both companies say they are separate. WorldClaw told Reuters it is an "independently owned and operated AI company," and its website says it is "not offered, managed, or controlled in any way" by World Liberty. But a WorldClaw spokesperson confirmed that Ryan Fang, World Liberty's head of growth, has been a "helpful external adviser and supporter of WorldClaw, particularly regarding USD1 adoption, partnerships, and expanding global access to AI services" — a role it called "strictly advisory." Both Trump sons have promoted WorldClaw on X, where Eric Trump called it "the future of finance."
The Restriction Lists, Precisely
The designations vary in severity, and none bar the models from U.S. commercial use. The Department of Defense designates Alibaba and Baidu as Chinese military-aligned companies — a listing that blocks the Pentagon from doing business with them but does not restrict private buyers. Z.ai, formerly Zhipu AI, sits on the Commerce Department's entity list, a sharper instrument: U.S. firms selling to a listed entity must seek an export license under a presumption of denial. The listing says Z.ai advances China's "military modernization through the development and integration of advanced artificial intelligence research." WorldClaw also carries DeepSeek and Moonshot models; officials have accused both of stealing U.S. IP.
The denials are on the record. "Alibaba is not a Chinese military company nor part of any military-civil fusion strategy," a spokesperson told Reuters, calling the Pentagon listing "arbitrary and capricious" and saying the company would sue for removal. Baidu and Z.ai also dispute the finding; Beijing disputes the theft claims. Baidu, Z.ai, DeepSeek and Moonshot did not respond to Reuters, nor did Commerce or Defense.
White House spokesperson Anna Kelly said "there are no conflicts of interest" in the relationship and that "President Trump only acts in the best interests of the American public." David Wachsman, a World Liberty spokesman, said WorldClaw is independent and that major U.S. firms likewise offer Chinese and American models: "This is a common and widely accepted approach." A WorldClaw spokesperson said the company "helps American AI companies reach international users" and that "making a model available does not constitute an endorsement of its developer."
WorldClaw's website says it may share user inputs with the companies supplying its models. Daniel Remler of the Center for a New American Security, until recently a State Department policy adviser, told Reuters the models could expose users to Chinese government monitoring, censored outputs, and injected code that hijacks AI agents. WorldClaw said it applies safeguards.
Why It Matters
Strip out the president and this is still a policy problem. Washington's AI statecraft is built for chokepoints — chips, entity lists, license denials — and has little purchase on model weights published openly and served by anyone who wants to. The economics do the rest. OpenRouter's June 2026 analysis priced DeepSeek V4 Flash at a weighted-average $0.054 per million input tokens and $0.242 output, and ranked Z.ai's GLM 5.2 the top open-weight model on the Artificial Analysis Intelligence Index. GLM 5.2 landed days after a U.S. export-control directive forced Anthropic to disable Fable 5 and Mythos 5 broadly to block foreign-national access — making an unrestricted near-frontier alternative more attractive, not less.
Aggregators are where that substitution goes invisible. WorldClaw says its WorldRouter tool serves more than 10,000 users and over 50 million requested tasks a day. A router abstracts provenance away: the buyer picks a price and a capability, not a country. Entity listings were built to stop exports to a company — not for a world where the product is a file and payment clears in stablecoin.
Peter Jeydel, who leads the sanctions and trade controls team at Troutman Pepper Locke in Washington, told Reuters the collaboration cuts both ways: "On one hand, people could say this is a China-hawk administration, and there is a tension with this type of business collaboration. On the other hand, it can be argued that this Trump presidency has been less aggressive in targeting Chinese companies in favor of a business-first approach."
What To Watch
Three threads. Whether Commerce or Defense answers the questions they declined to address; extending entity-list logic to model-serving intermediaries would be a real escalation. The Trump–Xi summit next month, which both governments have approached by trying to lower AI tensions — a détente cuts against new designations. And Congress: World Liberty won conditional OCC approval for a national trust bank charter on August 14, prompting Senator Elizabeth Warren to call it "the most brazen act of self-dealing our financial system has ever seen." That fight is about banking, not AI — but it is the same holdings and the same committee.
“As the U.S. government tries to respond to the rise and threat of Chinese AI, it seems hypocritical to go out through WorldClaw to use these tools from China to try and make a bunch of money.”— Sam Bresnick, Fellow, Georgetown Center for Security and Emerging Technology