Skip to main content
Explainer

Do we have to label AI-generated content from 2 December 2026?

Not from December: the labelling duties in Article 50 of the EU AI Act have applied since 2 August 2026. If you publish a deepfake, or AI-written text on a matter of public interest without editorial review, you must disclose it now. 2 December 2026 is a grace period for AI vendors, not for you.

4 min read

Do we have to label AI-generated content from 2 December 2026?

No. For most organisations the date is already behind you.

The labelling duties in Article 50 of the EU AI Act have applied since 2 August 2026. If your team publishes a deepfake, or AI-written text that informs the public on a matter of public interest without human editorial review, you have had to say so since that day.

The 2 December 2026 date is real, but it is narrow. It gives the companies that build generative AI tools four extra months to watermark the output of systems they had already put on the market. It does not give a marketing team, a newsroom or a webshop more time.

Who carries which duty?

The AI Act splits the work between two roles. The provider develops an AI system and puts it on the market under its own name: the company that sells the image generator. The deployer uses a system under its own authority: your marketing team when it uses that generator for a campaign (Article 3(3) and (4)).

That split decides almost everything.

Machine-readable marking is the provider's job; telling your audience that a video is fake is yours. Personal, non-professional use is outside the deployer definition, so an employee trying a tool at home is not the point. The campaign is. A team that buys a tool from a provider which has not yet added watermarking still owes its own disclosure, because the two duties run independently.

What Article 50 requires, and from whom

DutyWhoApplies from

Tell people they are talking to an AI, unless that is obvious (Art. 50(1))

Provider

2 August 2026

Mark synthetic audio, image, video and text in a machine-readable, detectable way (Art. 50(2))

Provider

2 August 2026; 2 December 2026 for systems on the market before 2 August 2026 (Art. 111(4))

Inform people exposed to emotion recognition or biometric categorisation (Art. 50(3))

Deployer

2 August 2026

Disclose that a deepfake is artificially generated or manipulated (Art. 50(4))

Deployer

2 August 2026

Disclose AI-generated text published to inform the public on matters of public interest, unless a person holds editorial responsibility after human review (Art. 50(4))

Deployer

2 August 2026

What counts as a deepfake?

A deep fake under the AI Act is image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear authentic or truthful. Text is never a deepfake; it has its own rule.

The two ends are easy. A photorealistic clip of your real CEO saying something she never said is a deepfake. A cartoon mascot is not. The middle is where teams get stuck: a realistic photo of a "customer" who does not exist, or of your own office with a product that is not on sale yet. The Commission's guidelines on Article 50, published on 20 July 2026 and updated on 31 July, are the place to start for those cases, and counsel is the place to finish.

Evidently artistic, satirical or fictional work gets a lighter rule: disclose only that generated content exists, in a way that does not spoil the work.

What about text written with AI?

The text rule is narrower than most people assume. It covers AI-generated text published to inform the public on matters of public interest, and it falls away when the text has gone through human review or editorial control and a person or company holds editorial responsibility for publishing it. In practice that means a named editor who reads and signs off removes the disclosure duty for an article. Article 3 does not define "matters of public interest", so product copy about your own range is a weak fit and commentary on elections, health or the economy is a strong one.

So what happens on 2 December 2026?

Regulation (EU) 2026/1744, the Digital Omnibus on AI, entered into force on 27 July 2026. It inserted a new Article 111(4): providers of generative AI systems placed on the market before 2 August 2026 must comply with the marking duty in Article 50(2) by 2 December 2026. Recital 38 calls this a transitional period of four months for providers. Deployers are not mentioned.

The same amending regulation adds two prohibitions that apply from 2 December 2026: AI systems that generate realistic intimate images of an identifiable person without consent, and systems that generate child sexual abuse material. Those are bans on the systems, not labelling rules.

What a marketing team can do this month

  • Art. 50(4)

    List what looks real

    Go through live and planned campaigns for AI-made images, audio or video that resemble real people, places or events. Those need a visible disclosure today.

  • Art. 50(4)

    Name the editor

    For AI-assisted articles on public-interest topics, record who reviewed the text and who carries editorial responsibility. That record is what takes the text outside the disclosure duty.

  • Art. 50(2)

    Ask your vendor

    Ask each AI tool supplier whether its output carries machine-readable marking, and from which date. If the tool was on the market before 2 August 2026, the answer may be 2 December.

What does getting it wrong cost?

Under Article 99, breaches of Article 50 can be fined up to EUR 15 million or 3% of worldwide annual turnover, whichever is higher. For SMEs and start-ups the cap is whichever of the two is lower.

In the Netherlands, the cabinet's draft implementing act makes the Autoriteit Persoonsgegevens the market surveillance authority for the transparency duties, with DNB and the AFM for financial institutions. The Commission's AI Office published a voluntary Code of Practice on marking and labelling on 10 June 2026.

Questions and answers

Does the four-month grace period cover our marketing team?

No. Article 111(4) is addressed to providers of AI systems and covers only the marking duty in Article 50(2). The deployer's duty to disclose deepfakes and public-interest text has applied since 2 August 2026.

Do we need to label every image we edited with AI?

No. Deployers disclose deepfakes, which are content that would falsely appear authentic. Separately, the provider's marking duty does not apply to tools that only perform an assistive function for standard editing, such as routine retouching.

Our agency makes the content with its own tools. Who is the deployer?

On the definitions, the deployer is whoever uses the AI system under its authority, which for the agency's own tools is the agency. Because you publish the result, put in the contract who discloses and how.

Is the Code of Practice binding?

No. The Commission describes it as voluntary. Under the amended Article 50(7), the Commission can adopt an implementing act with common rules if it judges the codes inadequate.

Regulations

  • EU Artificial Intelligence Act (EU AI Act)

    The EU AI Act is Regulation (EU) 2024/1689. On 2 August 2026 its transparency rules and enforcement machinery took effect, but the high-risk regime for credit scoring and life and health insurance pricing did not: Regulation (EU) 2026/1744 moved it to 2 December 2027. Prohibitions have applied since February 2025.

    Next
    New Art. 5 prohibitions; Art. 50(2) marking deadline for pre-existing generative systems

    Checked