China has operated a mandatory registration framework for large language models (LLMs) since 2023, while US President Donald Trump last week abandoned plans to require federal agencies to review advanced AI models — citing the need to preserve America's competitive edge over Beijing. The divergence marks a pivotal moment in how the world's two largest AI powers are approaching governance of the technology.

Trump pulls back on AI oversight

President Trump scrapped a draft executive order that would have directed government agencies to conduct safety reviews of advanced AI models. '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, according to CNBC. He added that he felt the order 'could have been a blocker.'

The decision reflects the Trump administration's broader posture: treat AI regulation as a competitive liability rather than a safety imperative. The White House reversal came even as both countries are reportedly considering opening a formal dialogue on the rapidly advancing technology.

How China's layered AI registry actually works

Under rules administered by the Cyberspace Administration of China (CAC) — the country's top internet regulator — AI model developers must submit filings before any public release. The process, known as generative AI services filing, requires companies to provide a security self-assessment report, a keyword interception list, and testing questions.

Documents are first reviewed at the provincial-level CAC, then escalated to the central CAC. The full procedure typically takes three to six months, according to a March blog post by AllBright Law Offices. Major developers subject to this framework include DeepSeek, Zhipu AI, Alibaba Group Holding, and Tencent Holdings.

A lighter track for AI applications

Firms deploying AI applications — such as AI agents that call the API of an already-approved third-party model — fall under a separate, lighter regime called generative AI services registration. This track does not require a security self-assessment and is reviewed only at the provincial level.

As of April 2026, the CAC had logged 868 filed generative AI services and 530 registered ones, according to figures the regulator publishes periodically.

Why it matters

The contrast is stark: China now has a functioning, quantifiable AI registry with nearly 1,400 approved or registered services, while the US has moved away from even a review mandate. Critics of deregulation warn that the absence of federal oversight could allow unsafe models to proliferate; proponents argue it accelerates innovation and keeps American firms globally competitive.

The debate is no longer purely domestic. With both governments reportedly exploring bilateral AI talks, each side's regulatory posture will shape the terms of any future agreement on standards, safety benchmarks, or export controls.

What's next

The immediate question is whether US tech companies — including firms such as Anthropic — will push for voluntary safety frameworks to fill the federal vacuum, or whether Congress will step in. On the Chinese side, the CAC registry continues to expand, and the pace of approvals will be a key indicator of how quickly domestic AI products can reach consumers. Watch for any joint communiqué from potential US–China AI dialogue sessions, which could set informal global norms even without binding regulation.