China's AI Filing Regime: The Algorithm Registration System That Gates Market Access

China requires every public-facing generative AI service to file with the Cyberspace Administration of China before launch — the mechanism that forced Apple to route Apple Intelligence in China through Alibaba's Qwen and Baidu rather than its own or Western models.

Created 2026-07-15 Last reviewed 2026-07-15

What it is

Since August 2023, any company that wants to offer a generative AI service to the Chinese public must first register it with the Cyberspace Administration of China (CAC), the country’s internet and content regulator. The rules — the Interim Measures for the Management of Generative Artificial Intelligence Services, issued jointly by the CAC, the National Development and Reform Commission, the Ministry of Education and several other agencies on July 10, 2023, and effective August 15, 2023 — require providers whose services have “public opinion attributes or social mobilization capabilities” (in practice, almost any consumer-facing chatbot, image generator or assistant) to complete a security assessment and file their underlying algorithm with the CAC before the service can go live.

In practice this created a two-track system. Companies that build and train their own models must complete a fuller “filing” (备案) covering algorithm security, training data provenance and content-moderation safeguards. Companies that simply deploy an already-filed model without altering its architecture can go through a lighter “registration” (登记) process. The CAC publishes batches of approved services on a public registry, updated periodically — by April 2025 the government reported 346 generative AI services had been filed nationally.

The regime has continued to expand into new product categories as the technology moves. On July 15, 2026, the CAC’s official “Cyberspace China” channel announced the first clearances specifically for on-device, smartphone-based AI models — seven services from Apple, Huawei, Samsung, Xiaomi, Oppo, Vivo and Nubia. This was a distinct milestone: it extended a filing regime built around cloud-hosted chatbots to AI features running locally on handsets, and it was the gate through which Apple Intelligence entered the Chinese market.

Why it matters for AI governance and narratives

The filing regime is the mechanism through which Beijing converts market access into a lever over how AI actually gets built for Chinese users. Apple could not simply ship the same Apple Intelligence stack it uses elsewhere; regulatory approval required pairing with domestically filed models — Alibaba’s Qwen reportedly handling language tasks and Baidu’s technology handling other functions, according to Chinese and Western press coverage of the clearance. A foreign frontier product only entered the market as what the observatory’s editorial calls a “compliance skin”: the visible interface remains Apple’s, but the generative capability underneath is domestically sourced and already vetted by the state.

This matters for the AI narrative contest in two ways. First, it is a case study in how content-and-security regulation functions as industrial policy: the filing requirement does not just police speech, it structurally requires foreign AI companies to route capability through Chinese partners, reinforcing domestic model providers’ position in the world’s largest smartphone market. Second, it produces exactly the kind of divergent framing the observatory tracks — coverage in Chinese state-aligned outlets emphasizes regulatory approval and successful compliance, while coverage in outlets like the South China Morning Post and Western tech press tends to foreground what the arrangement reveals about the limits of foreign AI sovereignty in China. Neither framing is neutral; both are strategic communications about who controls the terms of AI deployment.

Key facts and dates

Coverage of the exact division of labor between Qwen and Baidu inside Apple’s China stack, and of the precise filing steps Apple completed, varies in detail across outlets; the underlying regulatory framework and the fact of the July 2026 mobile-AI clearance are corroborated across multiple independent sources.

Where to learn more

Sources

Primary source: the official translated text of the regulation itself, from China Law Translate, the most widely cited translation source for Chinese legal texts.
Library of Congress Global Legal Monitor — authoritative government legal-research summary of the measures' promulgation and effective date.
Official Chinese State Council Information Office release confirming registry scale and ongoing operation of the filing system.
Caixin Global, a leading independent Chinese business/policy outlet, reporting the July 2026 on-device AI clearance and the Alibaba/Baidu partnership details.
Referenced in: Editorial No. 230