China is trying to turn AI content labeling from a voluntary product feature into a legal responsibility. A draft regulation from China’s Cyberspace Administration would require AI companies and online platforms to help users understand when images, audio, video, text or virtual-reality scenes were created by generative AI.
The proposal targets a practical problem: as AI systems improve, it becomes harder for ordinary users to tell whether a piece of content is real or machine-made. The source article links that confusion to risks ranging from nonconsensual porn to political disinformation.
What China Wants AI Companies To Label
On September 14, China’s Cyberspace Administration drafted a new regulation focused on labeling AI-generated content. The core idea is simple: people should be told when they are looking at, listening to or interacting with material created by AI.
The draft regulation asks AI service providers to use explicit labels. Those labels could take different forms depending on the medium.
- Images could carry visible watermarks.
- AI-generated video or virtual reality scenes could begin with “conspicuous notification labels”.
- AI-generated audio could include the morse code of “AI” (· – · ·) before or after the clip.
Some of these practices already exist in the industry. The important shift is that the Chinese proposal would make them legal obligations rather than optional trust-and-safety choices.
That change raises the stakes for AI developers and platforms. If labeling systems are weak, incomplete or missing, companies could face government penalties under the proposed approach.
Why Visible Watermarks Are Not Enough
The draft does not stop at visible labels. That is because visible marks can often be removed or changed. A watermark can be cropped out of an image. A label at the end of a video can be edited away.
To address that weakness, the proposal also calls for implicit labels in file metadata. According to the source article, the metadata should include the initialism “AIGC” and encrypted information about the companies that produced and spread the file.
The regulation also recommends invisible watermarks. These would not be obvious to users, but they could help identify AI-generated content after it has moved across tools and platforms.
This is where the challenge becomes much larger than adding a badge to a post. Metadata only works well if many companies follow common rules. AI model providers, deployment tools, editing software and social platforms would all need to preserve and read the same kinds of signals.
“Interoperable standards for metadata require that they work across AI models and deployers, tools and platforms—that's a tall order and does have both technical challenges and costs for the change,” says Sam Gregory, the executive director of Witness, a human rights organization in New York.
Gregory says that kind of system would take years, not months. The draft regulation therefore points to a future in which AI labeling is not just a design choice, but a shared technical infrastructure problem.
Social Platforms Would Carry A Bigger Burden
The most significant part of the Chinese proposal may be its treatment of social media and other online services. The draft asks “online information content transmission platform services” to examine shared files for implicit labels and signs of AI generation.
Platforms would need to add a generative AI tag or label in several cases: when metadata indicates AI use, when the uploader voluntarily says the content is AI-generated, or when the platform suspects that it is AI-generated.
They would also need to add their own information into metadata, creating a record of how the content moved online. That would make platforms part of the labeling chain, not just places where labeled content appears.
Jay Si, a Shanghai-based partner at Zhong Lun Law Firm, said there is still uncertainty over which services would count. Social media platforms like Douyin, WeChat, and Weibo are likely covered, but it is unclear whether ecommerce platforms like Taobao and JD or search engines like Baidu would also fall under the rule.
Some popular vertical video platforms in China already allow users to mark a post as AI-generated. Some also let users flag untagged videos or apply a notice saying, “The content is suspected of being generated by AI.”
A legal duty to screen content would be different. Platforms with hundreds of millions of users inside and outside China would need the technical ability and staffing to check huge volumes of uploads.
“If WeChat or Douyin needs to examine every single photo uploaded to the platform and check if they are generated by AI, that will become a huge burden in terms of workload and technical capabilities for the company,” Si says.
Douyin and Kuaishou declined to comment for the source story.
How The Proposal Compares With Other AI Rules
China is not alone in pushing AI labels. The European Union’s AI Act, adopted this March, also requires similar labels. California passed a similar bill this month. China’s earlier AI regulations had also briefly mentioned the need for generative AI labels.
What makes this draft stand out is the level of detail and the possibility of penalties for platforms where AI-generated content spreads without proper classification.
Angela Zhang, a law professor at the University of Southern California studying Chinese tech regulations, said China is moving ahead in AI content moderation partly because of the government’s demand to ensure political alignment in chatbot services. She also said labeling is an area where global technical standards may be possible.
The EU AI Act requires that “outputs of the AI system are marked in a machine-readable format and detectable as artificially generated or manipulated.” It also calls for explicit disclosure when content includes deepfake visuals or text information involving public interests.
Some companies have started using the C2PA standard, a metadata-based provenance approach. Gregory said supporters of Coalition for Content Provenance and Authenticity (C2PA) include Google, Meta, Microsoft, and OpenAI. He described it as progress, while noting that it is not yet widely available and that many platforms have not adopted it.
Jeffrey Ding, an assistant professor of Political Science at George Washington University, said Chinese regulators likely learned from the EU AI Act. At the same time, he noted that requiring platforms to screen user-uploaded content for AI may be specific to China’s context.
What Happens Next
The draft regulation is seeking public feedback until October 14. After that, it may take another several months for the proposal to be modified and passed.
Even before it takes effect, the direction is clear. China wants AI-generated content to carry both visible signals for users and hidden signals for machines. It also wants platforms to help detect, label and trace that content as it spreads.
For AI companies and social platforms, the message is that labeling is no longer just a transparency feature. Under this proposal, AI watermarks, metadata and platform checks could become part of the cost of operating in China’s online information environment.