New regulations on artificial intelligence take effect starting September 1.

Introduction

In recent years, generative artificial intelligence technology has been developing rapidly. AI-generated text, images, audio, and video are deeply integrated into all aspects of social life. From intelligent customer service to virtual anchors, and from creative design to medical diagnosis, AI-generated content is driving societal progress with astonishing creativity. However, behind these technological benefits lie hidden challenges: issues such as fake news, forged audio and video, and infringing content are becoming increasingly prevalent, posing a serious threat to the security of the online ecosystem. To address these challenges, the Cyberspace Administration of China, in collaboration with multiple government departments, has issued the “Measures for Identifying AI-Generated Synthetic Content” (hereinafter referred to as the “Identification Measures”), encouraging service providers to add implicit identifiers—such as digital watermarks—in their AI-generated synthetic content. If a user requests a service provider to deliver AI-generated synthetic content without explicit identifiers, the service provider may, after clearly stipulating the user’s identification obligations and usage responsibilities in the user agreement, provide such content without explicit identifiers and, in accordance with the law, retain relevant logs—including information on the recipients—for no less than six months.

Core of the new regulation: AI-generated content must “reveal its identity.”

The “Signage Measures” are a pioneering initiative. The “visible and invisible dual-identification” system builds a closed-loop governance framework for human-machine collaboration.

Explicit identification: All AI-generated text, images, audio, and video must be prominently labeled on the interface with the words or symbol “AI-generated,” enabling the public to instantly distinguish between real and AI-generated content and granting users the right to know and the right to choose.

Implicit identifier: By employing technologies such as digital watermarking and metadata embedding, an unalterable mark is embedded at the file’s underlying layer. “Stealth ID,” ensuring that falsified content is traceable and accountable.

This kind of By combining “visible reminders” with “invisible safeguards,” we can both ensure the safety of public information consumption and provide regulatory authorities with a tool for penetrating, real-time oversight.

Responsibility Loop: Fully Implementing Principal Responsibility Across the Entire Chain  

The “Identification Measures” clearly define the boundaries of responsibilities for all parties:  

The new regulations clearly define the obligations of three types of entities:

Service provider: The obligation to provide clear labeling must be strictly enforced. For high-risk content in scenarios such as news and finance, explicit labels must be mandatorily added, and implicit labels must be embedded in the metadata.  

Distribution platform: A technical filtering mechanism needs to be established to block the dissemination channels of content that is either unmarked or improperly marked.  

Individuals and Organizations: It is strictly prohibited to tamper with or forge labels, and no technical tools may be provided for malicious activities. Violators will be held accountable according to law.  

Through Strengthen accountability across the entire “generation—dissemination—use” chain, and promote the establishment of a governance framework characterized by “integration of technology and management, and multi-party collaborative governance.”

Moving Forward with Renewal: Safeguarding Innovative Vitality Through Order

The issuance of the “Labeling Measures” is by no means a move to... Rather than “slamming on the brakes” in AI development, we should address the trust crisis of “distinguishing between truth and falsehood” by establishing a credible identification mechanism.

For the public, explicit labeling is... “Anti-Fraud Guide”—the implicit label is “Rights-Protecting Shield”;

For businesses, unified standards reduce compliance costs, while a trusted environment unlocks commercial potential.  

For industries, clear regulatory boundaries will guide resources toward high-quality innovation pathways.  

As the head of the National Internet Information Office stated: “Standardization is for better development. We must not only be vigilant about technological risks but also fully harness the benefits of innovation.”

As The official implementation of the "Labeling Measures" on September 1, 2025, marks a crucial step forward for China in the field of AI governance. This practice—governing technology with technology—not only contributes Chinese wisdom to global digital governance but also underscores a shared consensus: Only when technological innovation and institutional innovation resonate in harmony can artificial intelligence truly become a new type of productive force that benefits humanity.