Can DeepSeek's mandatory AI tagging on WeChat and Douyin prevent fraud?
In the past few years, AI-generated content has flooded into our lives. However, AI often "hides" behind the scenes. This "invisible presence" will come to an end starting September 1st.
Whether it's a video you scroll across on Douyin, a science popularization animation on Bilibili, a live e-commerce stream on Kuaishou, or even a paragraph of text in a news push, you must be informed whether it is AI-generated. If it is AI-generated but not marked as such, it is a violation.
Through browsing major content platforms, Zhidx found that a large amount of AI-generated and synthesized image and video content has been marked as "AI-generated" or "content suspected to be AI-generated". Meanwhile, many text-based AI-synthesized contents are not marked, and it is still difficult to identify this type of content.
▲ Markings of AI-generated content on Xiaohongshu (left), Douyin (middle), and QQ Music (right)
Starting September 1st, the Measures for the Marking of AI-Generated and Synthesized Content (hereinafter referred to as the "Measures") jointly issued by the Cyberspace Administration of China, the Ministry of Industry and Information Technology, the Ministry of Public Security, and the National Radio and Television Administration officially came into effect. The Measures clearly require that all AI-generated text, images, audio, videos, and virtual scene content must carry explicit or implicit markings and can no longer "mingle" with content created by humans.
▲ Partial screenshot of the announcement of the "Measures"
The mandatory national standard "Network Security Technology - Methods for Marking AI-Generated and Synthesized Content" was implemented simultaneously with the "Measures" on September 1st. It establishes a closed-loop mechanism for review, monitoring, and accountability at the regulatory level and standardizes the marking system at the technical level.
As the Measures were launched, major mainstream content platforms such as Tencent, Xiaohongshu, Sina Weibo, Douyin, Kuaishou, and Bilibili, as well as large model manufacturers such as DeepSeek, Tencent (Yuanbao), Baidu, Zhipu Qingyan, MiniMax, SenseTime, and Kunlun Wanwei have all voiced their support, announcing that they have fully implemented the new regulations. AI is no longer an "invisible person" behind the scenes but will coexist with us in a visible and traceable way.
All content platforms have proposed voluntary declarations by creators and compliant markings by platforms. Large model service providers mainly propose to add explicit and implicit markings to all generated content and guide and remind users not to delete or forge markings. Some content platforms and large model manufacturers have also issued penalty regulations such as content removal, traffic restrictions, and permanent bans.
However, there are still some problems and controversies in the initial implementation of the policy. While many authors support and voluntarily mark AI-generated and synthesized content, some creators believe that marking AI content should not be a one-size-fits-all approach. They suggest fine-grained management based on the degree of AI involvement. Some users have also been "wrongly affected". Original content has been misidentified as AI-generated, and appeals have been ineffective, or users have been restricted in traffic even after voluntarily marking their content. These are areas where the "three-tier detection" mechanism in the implementation of the "Measures" still needs improvement.
In the long run, for users, this may be a major change in the information ecosystem; for platforms and manufacturers, it is a major compliance test.
01 Three Days After the Implementation of the "Measures": Some Support to Prevent Fraud, While Others Complain About Wrongful Affects and Traffic Restrictions
The "Measures" clarify that AI-generated and synthesized content refers to information such as text, images, audio, videos, and virtual scenes generated and synthesized using AI technology. This content must be distinguished through both explicit and implicit markings.
Explicit markings: Such as text watermarks, corner marks, and voice prompts, which directly remind users that the content is AI-generated. Implicit markings: Hidden in the file through technical means such as metadata and digital watermarks for content traceability and post-event accountability.
Zhang Jiyu, the executive dean of the Institute of Future Rule of Law at Renmin University of China, pointed out that the "Measures" have set up a "three-tier detection" mechanism:
1. If the content can be confirmed as AI-generated through implicit markings in the metadata, it should be directly marked.
2. If it can only be inferred through algorithms, it should be marked as "suspected to be AI-generated" to avoid wrongly affecting original content.
3. At the same time, experts suggest that platforms should establish appeal channels to allow users to submit evidence for review and safeguard the rights and interests of creators.
At the industrial level, experts point out that if unmarked AI content flows into public data sets, it will cause "data pollution", affecting the reliability of large model training. The "Measures" eliminate this risk at the source. Meanwhile, mandatory markings also help protect the traditional content industry, reduce infringement and fraud, and lower the compliance costs of platforms.
Once the "Measures" were implemented, they were widely supported by users. Many users believe that the rights and interests of original authors can finally be protected. At the same time, many users suggest that platforms add a "one-click block AI" function, stating that they are visually fatigued after seeing too much AI content and hope to see purely human-created content.
▲ User feedback on the Xiaohongshu platform as an example (Source: Xiaohongshu official account)
At the same time, some users have also raised objections. Some users believe that marking AI content should not be a one-size-fits-all approach. In some works, the proportion of AI content is less than 50%. They suggest that platforms introduce more detailed management methods to avoid discouraging those who truly want to use AI for creation. Some users also think that the marking watermarks are as ugly as stickers and hope for other ways to mark the content.
▲ User feedback on the Xiaohongshu platform as an example (Source: Xiaohongshu official account)
In addition, some users have reported that there have been cases of "wrongful affects" in content platforms in the early stages of policy implementation. Some users claim that their notes do not use any AI-generated content and are all real photos taken with mobile phones, but they have still been marked as "notes containing AI-generated content". Some users have complained, "Why am I restricted in traffic for a month just because of the marking?" and "I've appealed countless times, but it's useless". It can be seen that it still takes time to improve the "three-tier detection" mechanism proposed by experts.
▲ User feedback on the Xiaohongshu platform as an example (Source: Xiaohongshu official account)
02 Mainstream Content Platforms Take Action: Violations May Lead to Traffic Restrictions, Content Removal, and Account Bans
As the "Measures" were implemented, major domestic content platforms quickly took action, issuing announcements to clarify marking rules. Some platforms also announced that they would take measures such as traffic restrictions, content removal, and account bans against those who do not voluntarily mark AI-generated content.
On August 29th, Xiaohongshu's official account released an announcement regarding the marking of AI-generated and synthesized content, proposing voluntary declarations by creators and compliant markings by the platform.
Users can add a label by clicking on Settings → Content Type Declaration → Checking "Note contains AI-synthesized content" on the posting interface. For AI-generated and synthesized content that is not voluntarily declared, the platform has the right to add explicit or implicit markings based on legal and regulatory requirements, and users are not allowed to maliciously delete, forge, or conceal relevant markings.
▲ Xiaohongshu's announcement
As shown in the following figure, when we enter Xiaohongshu, we can see that many pieces of content have been marked as AI-generated.
▲ Voluntary markings by users (left) and markings identified by the platform (right) appear in different positions
However, when it comes to some AI-generated text content, such as when some bloggers clearly state that the content is an answer from DeepSeek, there are still no automatic markings or platform-identified markings on the content page.
▲ No markings on the text content generated by DeepSeek
On the same day, Sina Weibo's official account announced that it would implement the AI marking work in accordance with the "Measures". Users need to make voluntary declarations. On the basis of adding an option for users to voluntarily declare AI content on the posting side, the platform has added a complaint category of "No AI marking added" on the complaint side. The platform will add implicit and explicit markings to AI-generated content identified on the platform in accordance with regulations.
▲ Sina Weibo's announcement
When logging into the Sina Weibo platform, Zhidx temporarily found that there are still many pieces of content involving AI-generated images posted by bloggers without markings.
▲ Many AI-generated contents on Sina Weibo have not been marked
Bilibili also launched the AI-generated content marking function in advance on the same day. UP owners can voluntarily declare "This video uses artificial intelligence synthesis technology" when submitting videos. Otherwise, the platform will automatically add markings in accordance with regulations. Bilibili said that it will continuously optimize the AI governance process and maintain community order.
▲ Bilibili's announcement
When we enter Bilibili, we can see that the markings declared by the author will appear below the content title.
▲ Screenshot of the marking of AI-generated content on Bilibili (Source: Bilibili account of Tuohydrogen Xiaomeimei)
On September 1st, multiple platforms under Tencent announced that they would add explicit and implicit markings to AI content. Tencent WeChat Public Platform, Tencent Music, and QQ Security Center have respectively issued detailed rules, requiring that all generated content must carry markings to ensure transparency and compliance.
