The EU AI Act, Article 50
Article 50 of the AI Act, Regulation (EU) 2024/1689, has applied since 2 August 2026 and splits the duties in two. Providers of generative systems must mark outputs in a machine-readable format so they are detectable as AI-generated; systems already on the market before that date have until 2 December 2026. Deployers, meaning organisations using AI systems professionally, brands and agencies included, must disclose deepfakes: image, audio or video that resembles people, objects, places or events that exist or could plausibly exist, and would falsely appear authentic. For evidently artistic, creative, satirical or fictional work the duty is lighter: disclose in a way that does not spoil the work. AI-generated text published to inform the public on matters of public interest must be disclosed, unless it has had human review and someone holds editorial responsibility.
China, South Korea and other markets
China's measures for labelling AI-generated and synthetic content took effect on 1 September 2025, alongside a national standard. They require explicit labels that people can see or hear, such as text on images, audio cues or notices on video, and implicit labels embedded in file metadata, and they oblige platforms to detect and label synthetic content that users upload. Brands publishing in China should expect platforms there to check for both. Elsewhere, rules are arriving market by market: South Korea's AI Basic Act, in force since 22 January 2026, requires businesses offering generative AI products or services to label their output, with fines deferred for at least a year, and other countries have proposals under way. Map the rules for every market a campaign runs in.
Platform labels and Content Credentials
Platforms add their own layer. Instagram, Facebook, TikTok and YouTube each ask creators to disclose realistic AI-made or altered content and apply labels of their own. All four say they read C2PA Content Credentials when labelling, and Meta's apps read IPTC metadata as well. Content Credentials are a signed record of how a file was made and edited, attached to the file and readable with a compatible viewer. Re-encoding and screenshots can strip them, which is one reason the EU code of practice on marking combines signed metadata with invisible watermarks, and why brands should pair credentials with a visible disclosure wherever the law or the platform requires one. Paid ads may carry separate advertiser disclosure rules, particularly for political and social issue content.
Realistic versus clearly stylised
The practical test is whether a reasonable viewer could mistake the content for a real recording of a real person, place or event. A photorealistic spokesperson who does not exist, a synthetic voice resembling a known presenter or a realistic scene of an event that never happened sits at the high end and needs clear disclosure. An obviously illustrated character, a stylised product world or an abstract animation sits at the low end, although machine-readable marking still applies to the output. Product imagery is its own case: a generated scene around a real product must still show the product truthfully. The studio logs the model behind every asset it produces, which gives each disclosure decision an evidence base.
Updated 25 September 2026 · General information, not legal advice. Rules change, so check the current text with your legal team before relying on it.