EU AI Act Article 50, from 2 August 2026
Article 50 of the EU AI Act — the transparency chapter — applies from 2 August 2026. The Commission's guidance on transparency of AI-generated content does not describe a transition or grace period beyond that date. If you run ads into the EU with AI-generated creative, this is the second disclosure regime landing on you inside two months, after New York's on 9 June.
What Article 50 actually says
The obligations split by role:
- Providers — design AI systems so people are explicitly informed when they are interacting with an AI system, and add machine-readable marks so AI-generated or manipulated content can be detected.
- Deployers — inform people when they are exposed to deepfakes, to emotion recognition or biometric categorisation systems, or to text published on matters of public interest without human review or editorial control.
Compliance can be shown through the Code of Practice on Transparency of AI-generated content, or through alternative means of equivalent adequacy.
An honest caveat. The Commission's transparency guidance does not set out specific requirements for AI-generated people in commercial advertising the way New York's statute does. Anyone telling you Article 50 is a drop-in equivalent of the synthetic performer law is overstating it. What is concrete is the date, the machine-readable marking duty, and the deepfake disclosure duty on deployers.
How it differs from New York's synthetic performer law
| New York GBL § 396-b | EU AI Act Article 50 | |
|---|---|---|
| In force | 9 June 2026 | 2 August 2026 |
| Who it binds | Advertisers running commercial ads | Providers and deployers of AI systems |
| Trigger | A synthetic performer — a figure that reads as human but is not any identifiable real person | AI interaction, AI-generated or manipulated content, deepfakes |
| Disclosure form | Conspicuous disclosure in the ad | Machine-readable marking, plus informing people for deepfakes |
| Penalty | $1,000 first violation, $5,000 after | Set by the AI Act's penalty regime |
The two regimes point in the same direction but do not overlap neatly. New York cares whether the figure on screen looks like a person who does not exist. Article 50 cares whether audiences can tell — by machine or by notice — that content was AI-generated or manipulated at all.
The platforms have already moved
You can see this converging in the ad platforms without waiting for enforcement. Since July 2026 Google has permitted labels drawn directly into image and video creatives, and for campaigns targeting the EU, India and New York it renders a visible overlay on assets designated as AI created or edited. In other words, Google is already treating these three jurisdictions as one operational bucket. Google also says explicitly that using its label setting “doesn't guarantee compliance with specific regulations”.
The practical read: the platform toggle is a tool, not a safe harbour. If a creative needs disclosure, put the disclosure where a person can see it, and keep a record of what you did.
A practical checklist for ad creative
- Inventory which creatives contain AI-generated humans, separating generic synthetic people from real people who were merely retouched. The distinction decides which regime applies.
- Keep the machine-readable mark. Metadata — XMP
dc:subject, C2PA and similar — is the machine-readable half. Amazon reads it today, and Article 50 pushes the same direction. Verify it survived after every export. - Add a visible label where the platform will not add one for you. Burn it into the creative, keeping it away from edges that get cropped.
- Turn off crop-happy enhancements — Google names Asset Enhancements and Adaptive Layouts as features that can trim rendered content.
- Record the provenance: which tool generated the asset, who reviewed it, which markets it ran in, and what disclosure was applied. Both regimes turn partly on what you knew.
Sources: the European Commission's guidelines on transparency of AI-generated content, and Google's advertising policy and Display & Video 360 documentation on AI content labels. This page summarises public documents and is not legal advice.