• Since 2 August 2026, Article 50 of the EU AI Act requires organizations using AI tools professionally to disclose deepfakes and to label AI-generated text published on matters of public interest, unless a qualified person reviewed it and someone holds editorial responsibility.
  • The 2 December 2026 date covers providers only: generative AI systems placed on the EU market before 2 August 2026 have until then to mark output in a machine-readable way. Deployer duties, which cover most marketing teams, did not move.
  • Content created before 2 August 2026 needs no retroactive label, and fines can reach 15 million euros or 3% of worldwide annual turnover.

Your team made a product video with an AI avatar for a campaign in Germany. Legal asks whether it needs a label, and the vendor says its tool “handles compliance”.

As of October 2026, the transparency rules in Article 50 of the EU AI Act have applied for almost three months. Below is what the European Commission and Morgan Lewis confirm, turned into a labeling checklist. This article is general information and does not constitute legal advice.

What does Article 50 of the AI Act require since 2 August 2026?

Article 50 has applied since 2 August 2026 and splits transparency duties between providers, who build AI systems, and deployers, who use them professionally. According to the European Commission’s Article 50 FAQ, providers must tell people when they are interacting with a chatbot, AI agent or avatar, unless that is obvious, and must mark synthetic audio, images, video and text in a machine-readable, detectable format.

Deployers must clearly label deepfakes and AI-generated or manipulated text published on matters of public interest without human review or editorial control, and must inform people exposed to emotion recognition or biometric categorization. Disclosure has to be clear, visible or audible without technical tools, and given at first exposure at the latest. The Commission puts maximum fines at 15 million euros or 3% of worldwide turnover for the preceding financial year.

What changes on 2 December 2026, and for whom?

Only providers of generative AI systems placed on the EU market before 2 August 2026 get until 2 December 2026, and only for machine-readable marking. Systems released from 2 August onward had to comply at once.

A Morgan Lewis post from 12 August 2026 states that the 2 December transition period does not postpone deployer duties. For marketing teams, labeling was due in August.

Is your marketing team a deployer or a provider?

In most cases your marketing team is a deployer, because it uses AI systems professionally, while the vendor that built the tool is the provider. Freelancers using AI for paid work count as deployers too.

The line can move: Morgan Lewis describes a deployer as using a provider’s system without substantial modification or white-labeling, so a chatbot you built or a vendor’s generator sold under your brand can make you a provider. Most teams source AI output from vendors such as AI video creation companies and AI tools for content marketing, so the marking duty sits with those vendors and the labeling decision sits with you.

For US and UK companies, the FAQ covers providers outside the EU whose output is used in the EU, and does not separately address deployers outside the EU, so check your position with counsel.

Which AI marketing assets need a label?

Label AI images, video and audio that would falsely appear authentic, and AI-written public-interest text that no qualified person reviewed. The Commission defines a deepfake as content that resembles existing persons, objects, places, entities or events and would falsely appear authentic, so realism is the test.

Asset Label needed? How
AI avatar or cloned voice resembling a real person Yes Visible or audible label from first exposure
Photorealistic AI image of a real product, place or event Likely, if it could pass as real Visible label on the asset
Evidently creative, satirical or fictional AI ad Yes, limited Caption or end card that does not spoil the work
Stylized illustration no one would take as real Probably not Record the reasoning
Unreviewed AI article on economic or other public-interest news Yes Label on the text
Same article after substantive human review No Keep a review record and a named responsible editor
AI chatbot on your site Provider duty, unless obvious Check the vendor’s disclosure
Emotion recognition in ad testing Yes, notify people Notice, also for after-the-fact analysis

For text, a documented human review is the main lever. The Code of Practice on Transparency of AI-generated Content is voluntary; companies that skip it must show compliance by other means and may face more information requests.

Frequently asked questions

Do I have to label every AI-generated image in EU ads?

No. The deployer duty targets deepfakes, meaning AI content that resembles real persons, objects, places or events and would falsely appear authentic. A clearly stylized illustration usually falls outside it, while a photorealistic scene or synthetic person usually falls inside. Evidently creative or satirical work still needs a light disclosure.

Does the 2 December 2026 deadline apply to marketing teams?

Generally no. The 2 December 2026 date covers only providers of generative AI systems placed on the EU market before 2 August 2026, and only their machine-readable marking. Deployer duties, including deepfake disclosure and public-interest text labels, have applied since 2 August 2026.

Does the EU AI Act apply to US or UK companies marketing in the EU?

For providers, yes: the Commission says providers outside the EU are covered when their system’s output is used in the EU. Its FAQ does not separately address deployers outside the EU, so a US or UK brand running AI campaigns for EU audiences should get advice on its position.

What counts as human review of AI-generated text?

Human review is a deliberate examination of the substance by a person with relevant knowledge and professional judgment. Spell-checking and grammar fixes do not qualify, and a person must hold ultimate legal responsibility for the publication. Reviewed text that meets those conditions needs no AI label.

Do we need to relabel AI content published before August 2026?

No. The Commission states that content generated before 2 August 2026 does not need retroactive labels. New assets fall under the rules, including new versions of old campaigns, so record when evergreen assets such as landing page videos were generated.

What should marketing teams do before 2 December 2026?

Audit EU-facing assets created since 2 August, add a human-review step with a named owner for AI-assisted text, and ask each AI vendor how it marks output. Morgan Lewis recommends getting this from providers, whose pre-August systems must carry marks by December. Teams that build the label decision into content approval make it once per asset, at sign-off.

Sources

Add Content Managers as a preferred source on Google