AI disclosure and labelling laws: EU, China, South Korea and the US

1. 1

   Tell people they are talking to AI

   Under EU AI Act Article 50, people must be told when they interact with an AI system unless it is obvious, from 2 August 2026. South Korea's AI Basic Act requires advance notice that a product or service runs on generative or high-impact AI. Utah requires disclosure of generative AI to consumers on request or in high-risk interactions, and up front in regulated occupations; Texas requires it from government agencies and health-care providers.

2. 2

   Mark synthetic content for machines

   EU Article 50(2) requires providers of systems that generate synthetic audio, images, video or text to mark the output in a machine-readable, detectable way, as far as technically feasible. China's labelling measures, in force since 1 September 2025, require implicit labels in file metadata alongside explicit ones.

3. 3

   Label deepfakes for people

   EU Article 50(4) requires deployers to disclose deepfakes and AI-generated text published on matters of public interest, with lighter rules for evidently artistic or satirical work. South Korea requires generative output to be labelled and realistic synthetic media to be clearly flagged. China requires visible labels on synthetic content.

4. 4

   Advertising and performers

   California SB 1050 requires a clear disclosure on ads that prominently include a synthetic performer, from 1 January 2027 (record pending review).

5. 5

   One design for all of them

   Use one provenance standard for machine-readable marking (metadata plus a watermark where content types allow), one visible label pattern per content type, a disclosure at the start of every AI conversation, and a register of the features that generate content. The synthetic content labelling plan template is built for this.

Obligations referenced in this guide

• Legal Disclose AI interaction and label synthetic content (EU AI Act, European Union)
• Legal Providers of generative AI must mark synthetic output as artificially generated in a machine-readable way (EU AI Act, European Union)
• Legal Deployers of emotion recognition or biometric categorisation must inform exposed persons (EU AI Act, European Union)
• Legal Deployers must disclose deepfakes and AI-generated text published on matters of public interest (EU AI Act, European Union)
• Legal Health-care providers must disclose the use of AI in patient services (Texas Responsible AI Governance Act (TRAIGA), Texas (United States))
• Legal AI business operators must label generative AI output and clearly flag realistic synthetic media (AI Basic Act, South Korea)
• Legal AI business operators must notify users in advance that a product or service runs on high-impact or generative AI (AI Basic Act, South Korea)

Frequently asked questions

Do I have to tell users they are talking to an AI?

In the EU, yes, under Article 50 of the AI Act from 2 August 2026, unless it is obvious. South Korea requires advance notice for generative and high-impact AI; Utah requires disclosure on request or in high-risk interactions, and Texas in government and health-care settings.

Do AI-generated images need a watermark?

The EU requires machine-readable marking of synthetic output (Article 50(2)), and China requires implicit labels in metadata plus explicit labels. A watermark is one technique; the laws on record ask for detectable marking rather than naming a single method.

What does China require for AI-generated content?

Its labelling measures, in force since 1 September 2025, require explicit labels that people can see and implicit labels embedded in file metadata for AI-generated and synthetic content.

When did EU Article 50 start to apply?

On 2 August 2026. Regulation (EU) 2026/1744 did not move it, according to the record.

Disclaimer: informational only, not legal advice. Verify every claim against the linked official sources and consult a qualified lawyer before acting.