# How China regulates generative AI

URL: https://ai-pulse.aglaganova.com/blog/how-china-regulates-generative-ai
Published: 2026-10-09
Author: Roman Aglaganov, Full-stack developer
Category: Policy
Region: Asia-Pacific
Tags: regulation, deepfakes

> China was among the first countries to write binding rules for chatbots and deepfakes. Providers must register their models, keep content in line with state values and, since September 2025, label AI-generated content.

![Typographic cover: Label it. AI-generated content rules, in force since 1 September 2025](https://ai-pulse.aglaganova.com/api/media/file/china-1.png)

While Europe spent years negotiating one big AI law, China moved faster with a series of targeted regulations, each aimed at a specific kind of AI. The result is one of the most detailed rulebooks in the world, with a distinctly Chinese emphasis on control of information.

## Rule by rule

**Recommendation algorithms (2022):** rules for the systems that decide what you see in feeds and apps, including a registry where companies file details of their algorithms with the regulator.

**Deepfakes (2023):** rules for “deep synthesis”, AI that generates or alters images, audio and video, requiring consent for editing someone’s likeness and visible marks on synthetic content.

**Generative AI (August 2023):** the Interim Measures for generative AI services, covering chatbots offered to the public. Providers must use lawfully sourced training data, protect personal information and ensure output upholds “core socialist values”. Public-facing models also go through a filing and security assessment process before launch.

**Labelling (September 2025):** rules requiring AI-generated text, images, audio and video to carry visible labels where appropriate, plus hidden labels in the file’s metadata.

## Two goals at once

The rules try to do two things at the same time: let Chinese companies compete at the frontier, and make sure AI does not produce content the state considers harmful or politically sensitive. In practice that means Chinese chatbots are both increasingly capable and noticeably reluctant to discuss certain topics.

**Why it matters elsewhere:** some ideas, such as mandatory labelling of synthetic media, are spreading worldwide; the EU AI Act has its own transparency rules for deepfakes. Others, like requiring models to reflect state values, are unlikely to travel.

## Sources

- [Interim Measures for the Management of Generative AI Services (China Law Translate)](https://www.chinalawtranslate.com/en/generative-ai-interim/)