The AI videos and AI images you encounter online—are they really AI-generated? By 2026, this question is no longer a technical discussion but a legal red line. Since September 1, 2025, the "Measures for the Labeling of AI-Generated and Synthesized Content" have officially taken effect—meaning all content produced using AI deep synthesis technology must carry an "AI-Generated" label.
1. Which Content Must Be Labeled
According to the Labeling Measures, the following three categories of scenarios require mandatory labeling:
Category 1: Publicly disseminated synthesized content
This includes AI-generated images or videos published on social media platforms, used in commercial advertising, or featured in news reports. For example, if you use Midjourney to generate e-commerce product images for ad placements, or use an AI digital human to create a promotional video posted on Douyin—these all fall within the scope of mandatory labeling.
Category 2: Deep synthesis involving real persons' likenesses or voices
If you use AI face-swapping technology to produce short videos, or use AI voice cloning to generate voiceover content, it must be clearly labeled in a prominent location. The compliance requirements for this type of content are stricter, as it involves legal risks related to portrait rights and voice rights.
Category 3: Information content that may influence public judgment
This includes AI-generated news scene images, political figure speech videos, and medical or health advice. If such content misleads the public, the consequences are severe, making labeling a baseline requirement.
2. Legal Consequences of Non-Labeling
According to Article 13 of the Labeling Measures, violations of labeling obligations are handled by relevant authorities including the Cyberspace Administration, telecommunications, public security, and broadcasting departments according to their respective duties. Specific penalties include:
Penalty 1: Rectification orders and warnings
First-time violations typically result in a rectification notice from the regulatory authority, requiring supplementary labeling or removal of the relevant content within a specified timeframe.
Penalty 2: Fines
For refusal to rectify or serious violations, fines ranging from 10,000 to 100,000 yuan may be imposed. For enterprise users, while this amount is not astronomical, it will leave an administrative penalty record that affects subsequent business qualification reviews.
3. Recommended Labeling Methods
There are currently three mainstream labeling methods in the industry:
Method 1: In-frame text labeling
Add the text "This content is AI-generated" in the corner of a video frame or at the bottom of an image. This is the most intuitive method, though it may affect visual quality.
Method 2: Metadata embedding
The C2PA (Coalition for Content Provenance and Authenticity) standard allows AI-generated labels to be embedded in file metadata. Users can view them through players or browsers that support this standard. This method does not affect visual aesthetics, but its adoption rate is not yet widespread enough.
Method 3: Platform-side tags
Platforms such as Douyin and Xiaohongshu have built-in AI content tagging features. Selecting the "Contains AI-generated content" option when uploading automatically adds the label. This is currently the most convenient compliance method.
4. Enterprise Compliance Checklist
First, establish an internal review process—all commercial materials generated using AI tools must undergo compliance checks before publication. Second, retain original creation records (including prompt instructions and generation process screenshots) for regulatory inspection. Third, stay updated on platform policy guidelines, as individual platform labeling requirements may be stricter than national regulations.
5. Common Misconceptions
Many people believe, "I only used AI to assist in editing an image, which doesn't count as deep synthesis, so I don't need to label it"—this understanding is incorrect. The Labeling Measures' criteria do not depend on how much AI technology you used, but on whether the final published content contains AI-generated elements. Even if you only used AI to swap a photo's background or adjust its color tones, as long as it is disseminated to the public, it is recommended to add a label.
