AI labeling laws worldwide
Laws and obligations for labelling AI-generated content in 17 jurisdictions. As of 11/10/2026.
In the EU, Article 50 of the AI Act has applied since 2 August 2026. Deployers must visibly label deep fakes, and providers of generative AI must mark their output in a machine-readable way. China (since 1 September 2025), India (since 20 February 2026), South Korea (since 22 January 2026) and Vietnam (since 1 March 2026) have their own labelling rules. The US has no federal law; California requires machine-readable provenance data since 2 August 2026, and New York a disclosure of AI performers in advertising.
Comparison
| Jurisdiction | Status | Applies from | Visible | Machine-readable | Who | Source |
|---|---|---|---|---|---|---|
| 馃嚜馃嚭 European Union | in force | 02/08/2026 | yes | yes | Providers, Deployers | P |
| 馃嚛馃嚜 Germany | in force | 02/08/2026 | yes | yes | Providers, Deployers | P |
| 馃嚘馃嚬 Austria | in force | 02/08/2026 | yes | yes | Providers, Deployers | P |
| 馃嚠馃嚬 Italy | in force | 02/08/2026 | yes | yes | Providers, Deployers | P |
| 馃嚜馃嚫 Spain | draft | open | yes | yes | Providers, Deployers | P |
| 馃嚚馃嚦 China | in force | 01/09/2025 | yes | yes | Providers, Platforms, Users | P |
| 馃嚠馃嚦 India | in force | 20/02/2026 | yes | yes | Providers, Platforms | P |
| 馃嚢馃嚪 South Korea | in force | 22/01/2026 | partly | partly | Providers | P |
| 馃嚮馃嚦 Vietnam | in force | 01/03/2026 | partly | yes | Providers, Deployers | S |
| 馃嚭馃嚫 California (USA) | in force | 02/08/2026 | no | yes | Providers, Platforms | S |
| 馃嚭馃嚫 New York (USA) | in force | 09/06/2026 | yes | no | Advertisers | P |
| 馃嚭馃嚫 United States (federal) | no law | open | no | no | P | |
| 馃嚡馃嚨 Japan | no law | open | partly | no | Users | S |
| 馃嚙馃嚪 Brazil | no law | open | partly | no | Advertisers | S |
| 馃嚚馃嚘 Canada | no law | open | no | no | P | |
| 馃嚞馃嚙 United Kingdom | no law | open | no | no | P | |
| 馃嚘馃嚭 Australia | no law | open | no | no | S |
Source: P = primary source, S = partly secondary sources only. Status refers to the labelling rule; in EU member states Article 50 AI Act always applies directly. As a dataset: gesetze.json
Recent changes
- 30/09/2026 路 California: SB 1000 takes effect. The visible disclosure option and the 1 million user threshold are gone; the latent disclosure remains. (per secondary source)
- 02/08/2026 路 EU: Article 50 AI Act applies. Deployers must label deep fakes visibly, with no transition period. In Germany, the Federal Network Agency has been in charge since the KI-MIG of 29 July 2026.
- 13/07/2026 路 Japan: parliament passes a labelling rule for realistic AI images in election campaigns, applying from 1 March 2027. (per secondary source)
- 12/06/2026 路 Spain: the implementing bill goes to parliament (BOCG A-97-1). A missing deep fake label would be a serious infringement. It has not been adopted yet.
- 29/05/2026 路 South Korea: the government defers fines under the AI Basic Act (in force since 22 January 2026) for at least one year, except in cases of serious harm.
- 01/03/2026 路 Vietnam: the AI law takes effect, with machine-readable marking by providers and labels for simulated real people. (per secondary source)
- 20/02/2026 路 India: a visible label and permanent metadata become mandatory for realistic AI images, video and audio (IT Amendment Rules 2026).
All jurisdictions
馃嚜馃嚭 European Union
Article 50 AI Act since 2 Aug 2026: deployers label deep fakes visibly, providers mark output machine-readably.
Readin force馃嚛馃嚜 Germany
Federal Network Agency in charge (KI-MIG since 29 Jul 2026), online complaints, fines under Art. 99 AI Act.
Readin force馃嚘馃嚬 Austria
Article 50 applies directly; Austria has not yet notified a market surveillance authority to the Commission.
Readin force馃嚠馃嚬 Italy
Article 50 applies directly, plus Law 132/2025 and a criminal offence for harmful deep fakes.
Readdraft馃嚜馃嚫 Spain
Article 50 applies; bill in parliament: AESIA in charge, a missing deep fake label would be a serious breach.
Readin force馃嚚馃嚦 China
Since 1 Sep 2025: visible label and metadata on all AI content, platforms check, removal banned.
Readin force馃嚠馃嚦 India
Since 20 Feb 2026: visible label plus metadata with identifier for realistic AI content; text excluded.
Readin force馃嚢馃嚪 South Korea
Since 22 Jan 2026: generative AI providers label output, deep fakes clearly recognisable. Fines deferred for a year.
Readin force馃嚮馃嚦 Vietnam
Per secondary source since 1 Mar 2026: providers mark machine-readably, deployers label misleading content.
Readin force馃嚭馃嚫 California (USA)
Since 2 Aug 2026 latent provenance data from AI providers; the visible notice was dropped by SB 1000.
Readin force馃嚭馃嚫 New York (USA)
Since 9 Jun 2026: ads with AI-generated people need a conspicuous notice. USD 1,000 or 5,000.
Readno law馃嚭馃嚫 United States (federal)
No federal labelling law; the FTC bans AI-generated testimonials. Rules in California and New York.
Readno law馃嚡馃嚨 Japan
No general duty. Per secondary source, realistic AI images in election campaigns need labels from 1 Mar 2027.
Readno law馃嚙馃嚪 Brazil
AI bill PL 2338/2023 pending; labelling so far only in election advertising (per secondary source).
Readno law馃嚚馃嚘 Canada
No federal law; the Artificial Intelligence and Data Act failed with Bill C-27 in 2025.
Readno law馃嚞馃嚙 United Kingdom
No AI act; the government relies on existing regulators, watermarking is voluntary.
Readno law馃嚘馃嚭 Australia
No duty; per secondary source, voluntary government guidance since November 2025.
ReadRead next
When AI images need a label
Article 50 AI Act in plain words, with the exceptions.
ReadDeadlinesWhen the obligation applies
2 August 2026 for retailers, 2 December 2026 for AI providers.
ReadPlatforms
Amazon, Etsy, Instagram and more
What platforms require and label themselves.
ViewTechnologyUnderstanding AI metadata
C2PA, IPTC and watermarks: what the file says.
Read
Label your first AI image
3 images right away, 30 a month with a free account, with a record for every image. On Shopify, the app labels your whole store.