AI-generated content labelling laws

Side-by-side comparison

European Union: Article 50 of the AI Act

The AI Act defines a deep fake as AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful ¹. Article 113 says the Regulation applies from 2 August 2026, with earlier dates for some chapters ¹. This page reflects the text as published in the Official Journal.

• Providers of AI systems that generate synthetic audio, image, video or text must mark outputs in a machine-readable format and make them detectable as artificially generated or manipulated ¹.
• The marking duty does not apply where the system performs an assistive function for standard editing or does not substantially alter the input ¹.
• Deployers of a system that generates a deep fake must disclose that the content is artificially generated or manipulated. For evidently artistic, creative, satirical or fictional work, disclosure is limited to an appropriate form that does not hamper the work ¹.
• Deployers that publish AI-generated text to inform the public on matters of public interest must disclose it. The duty does not apply where the content has undergone human review or editorial control and a person holds editorial responsibility ¹.
• The AI Office is to encourage codes of practice on detection and labelling, and the Commission may adopt implementing acts ¹.

China: the 2025 labelling measures

The Cyberspace Administration of China, the Ministry of Industry and Information Technology, the Ministry of Public Security and the National Radio and Television Administration jointly issued the measures, effective 1 September 2025 ². They define explicit labels as text, sound or graphics added to content or interfaces that users can clearly perceive, and implicit labels as marks added to file data by technical means that users do not readily perceive ².

Service providers must add implicit labels in file metadata, and platforms that spread content must take technical measures to regulate how generated content circulates ². A mandatory national standard on labelling methods took effect on the same day ².

California: the AI Transparency Act

The law defines provenance data as information about the origin of digital content and its history of modifications, in a format compliant with widely adopted specifications from an established standards-setting body ³. Sections were amended by AB 853 in 2025 and SB 1000, effective 30 September 2026 ³.

• A covered provider must offer a free disclosure verification tool that lets users check whether image, video or audio content was created or altered by its system and outputs any system provenance data detected ³.
• A covered provider must, where technically feasible, include a latent disclosure conveying the provider's name and the name and version of the system, which is permanent or extraordinarily difficult to remove ³.
• From 1 January 2027, a large online platform, defined by more than 2,000,000 unique monthly users, must detect compliant provenance data and let users inspect it ³.
• From 1 January 2028, capture device makers must offer latent disclosures in captured content and embed them by default ³.
• The Attorney General, a city attorney or a county counsel may sue for $5,000 per violation ³.

Where provenance standards fit

None of the three texts names a single technical standard. The C2PA specification is one standard for signed provenance records, and it says an asset can become separated from its manifest through removal or corruption of metadata ⁴. That is why label rules and anti-stripping rules appear together. For the type of content, the IPTC Digital Source Type vocabulary gives a shared term, trainedAlgorithmicMedia, for media created by generative AI ⁵.