Trump scrapped a federal AI model review order citing competition with China, while Beijing's mandatory AI registry has processed every major domestic model since 2023.
Trump last week killed a draft executive order that would have directed U.S. government agencies to conduct safety reviews of advanced artificial intelligence models before deployment. The stated rationale was competitive.
"We're leading China, we're leading everybody, and I don't want to do anything that is going to get in the way of that lead," Trump told reporters, as reported by South China Morning Post. He called the proposed order a potential "blocker." Notably, China, the rival Trump cited to justify deregulation, had already implemented a mandatory model registration system three years earlier without visibly losing competitive ground.
That juxtaposition frames the sharpest policy divergence in the current AI race. On one side: a U.S. government with no formal pre-deployment review process for frontier AI. On the other: a Chinese system that has processed filings from every major domestic model released since 2023.
How China's registry works
China's Cyberspace Administration (CAC), the country's top internet regulator, runs a multilayered filing requirement for any company seeking to publicly launch a large language model or AI service. Developers including DeepSeek, Zhipu AI, Alibaba, and Tencent must submit a security self-assessment, a keyword interception list, and a behavioral test question set before release.
Documents move first to the provincial CAC, then to the central CAC in Beijing. From submission to clearance typically takes three to six months, according to a March analysis from AllBright Law Offices cited by South China Morning Post. Companies building AI agents or applications on top of registered foundation models face a second, separate filing requirement for their own products.
Running since 2023, the registry now holds documentation on every significant Chinese model that reached users, including DeepSeek's releases, which drew international attention earlier this year. Critics note the system functions partly as a censorship mechanism. It also, however, generates a de facto audit trail of behavioral testing data, something no equivalent U.S. process now produces.
The deregulation case and its limits
This argument follows a broader pattern in how the current administration approaches emerging technology. What makes the AI case distinctive is using China as the explicit justification for less oversight rather than more. The logic, that matching a rival's governance structure would slow you down relative to that rival, is novel and largely untested.
Forbes reported in January that OpenAI reached $20 billion in annual recurring revenue and 800 million monthly active users, figures the administration can point to as evidence that a light-touch environment produces results. Scale and safety are different dimensions, though. The scrapped order targeted risk evaluation, not market access.
Meanwhile, the two governments are reportedly considering a bilateral dialogue on artificial intelligence policy. Any formal conversation arrives with an asymmetry already built in: China brings three years of operational registry data and a tested compliance infrastructure; the U.S. brings a deliberate decision not to build the same.
What it means for developers
For teams operating across both markets, the divergence creates bifurcated compliance requirements. U.S. launches currently face no federal pre-release gate. Chinese launches require months of documentation, provincial review, and central agency clearance. Digital Watch Observatory has tracked the mounting commercial pressure on AI companies to demonstrate sustainable revenue models alongside rapid growth, and regulatory unpredictability adds real cost to cross-market operations.
If bilateral AI governance talks do advance, Washington will need to decide whether to build some version of what it just declined to build, or negotiate from a position of deliberate minimalism. Both carry long-term consequences that will take years to measure.
---
FAQ
What did Trump's scrapped AI executive order propose?
The draft order would have required federal agencies to review advanced AI models for safety risks before deployment. Trump said it risked slowing U.S. progress against China and shelved it.
How does China's AI model registry work?
Developers submit security self-assessments, keyword filter lists, and behavioral test questions to provincial regulators, then the central Cyberspace Administration of China. Approval typically takes three to six months.
Does the Chinese filing requirement cover AI apps, or just foundation models?
Both. Foundation model developers file first. Companies building AI agents or applications on top of those models must file separately for their own products.
Are the U.S. and China planning joint AI policy talks?
Reports indicate both sides are considering dialogue, but no formal process has been announced. The gap in regulatory infrastructure between the two countries will shape any eventual negotiation.